1 /*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
30 */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34
35 #include "opt_ipfw.h" /* for ipfw_fwd */
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_mac.h"
40 #include "opt_tcpdebug.h"
41
42 #include <sys/param.h>
43 #include <sys/kernel.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/proc.h> /* for proc0 declaration */
47 #include <sys/protosw.h>
48 #include <sys/signalvar.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sysctl.h>
52 #include <sys/syslog.h>
53 #include <sys/systm.h>
54
55 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
56
57 #include <vm/uma.h>
58
59 #include <net/if.h>
60 #include <net/route.h>
61
62 #define TCPSTATES /* for logging */
63
64 #include <netinet/in.h>
65 #include <netinet/in_pcb.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/in_var.h>
68 #include <netinet/ip.h>
69 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
70 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
71 #include <netinet/ip_var.h>
72 #include <netinet/ip_options.h>
73 #include <netinet/ip6.h>
74 #include <netinet/icmp6.h>
75 #include <netinet6/in6_pcb.h>
76 #include <netinet6/ip6_var.h>
77 #include <netinet6/nd6.h>
78 #include <netinet/tcp.h>
79 #include <netinet/tcp_fsm.h>
80 #include <netinet/tcp_seq.h>
81 #include <netinet/tcp_timer.h>
82 #include <netinet/tcp_var.h>
83 #include <netinet6/tcp6_var.h>
84 #include <netinet/tcpip.h>
85 #include <netinet/tcp_syncache.h>
86 #ifdef TCPDEBUG
87 #include <netinet/tcp_debug.h>
88 #endif /* TCPDEBUG */
89
90 #ifdef IPSEC
91 #include <netipsec/ipsec.h>
92 #include <netipsec/ipsec6.h>
93 #endif /*IPSEC*/
94
95 #include <machine/in_cksum.h>
96
97 #include <security/mac/mac_framework.h>
98
99 static const int tcprexmtthresh = 3;
100
101 struct tcpstat tcpstat;
102 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
103 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
104
105 int tcp_log_in_vain = 0;
106 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
107 &tcp_log_in_vain, 0, "Log all incoming TCP segments to closed ports");
108
109 static int blackhole = 0;
110 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
111 &blackhole, 0, "Do not send RST on segments to closed ports");
112
113 int tcp_delack_enabled = 1;
114 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
115 &tcp_delack_enabled, 0,
116 "Delay ACK to try and piggyback it onto a data packet");
117
118 static int drop_synfin = 0;
119 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
120 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
121
122 static int tcp_do_rfc3042 = 1;
123 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
124 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)");
125
126 int tcp_do_rfc3390 = 1;
127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
128 &tcp_do_rfc3390, 0,
129 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
130
131 static int tcp_insecure_rst = 0;
132 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
133 &tcp_insecure_rst, 0,
134 "Follow the old (insecure) criteria for accepting RST packets");
135
136 int tcp_do_autorcvbuf = 1;
137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
138 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
139
140 int tcp_autorcvbuf_inc = 16*1024;
141 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
142 &tcp_autorcvbuf_inc, 0,
143 "Incrementor step size of automatic receive buffer");
144
145 int tcp_autorcvbuf_max = 256*1024;
146 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
147 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
148
149 struct inpcbhead tcb;
150 #define tcb6 tcb /* for KAME src sync over BSD*'s */
151 struct inpcbinfo tcbinfo;
152
153 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
154 static void tcp_do_segment(struct mbuf *, struct tcphdr *,
155 struct socket *, struct tcpcb *, int, int);
156 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *,
157 struct tcpcb *, int, int);
158 static void tcp_pulloutofband(struct socket *,
159 struct tcphdr *, struct mbuf *, int);
160 static void tcp_xmit_timer(struct tcpcb *, int);
161 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
162
163 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
164 #ifdef INET6
165 #define ND6_HINT(tp) \
166 do { \
167 if ((tp) && (tp)->t_inpcb && \
168 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
169 nd6_nud_hint(NULL, NULL, 0); \
170 } while (0)
171 #else
172 #define ND6_HINT(tp)
173 #endif
174
175 /*
176 * Indicate whether this ack should be delayed. We can delay the ack if
177 * - there is no delayed ack timer in progress and
178 * - our last ack wasn't a 0-sized window. We never want to delay
179 * the ack that opens up a 0-sized window and
180 * - delayed acks are enabled or
181 * - this is a half-synchronized T/TCP connection.
182 */
183 #define DELAY_ACK(tp) \
184 ((!tcp_timer_active(tp, TT_DELACK) && \
185 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
186 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
187
188
189 /*
190 * TCP input handling is split into multiple parts:
191 * tcp6_input is a thin wrapper around tcp_input for the extended
192 * ip6_protox[] call format in ip6_input
193 * tcp_input handles primary segment validation, inpcb lookup and
194 * SYN processing on listen sockets
195 * tcp_do_segment processes the ACK and text of the segment for
196 * establishing, established and closing connections
197 */
198 #ifdef INET6
199 int
200 tcp6_input(struct mbuf **mp, int *offp, int proto)
201 {
202 struct mbuf *m = *mp;
203 struct in6_ifaddr *ia6;
204
205 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
206
207 /*
208 * draft-itojun-ipv6-tcp-to-anycast
209 * better place to put this in?
210 */
211 ia6 = ip6_getdstifaddr(m);
212 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
213 struct ip6_hdr *ip6;
214
215 ip6 = mtod(m, struct ip6_hdr *);
216 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
217 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
218 return IPPROTO_DONE;
219 }
220
221 tcp_input(m, *offp);
222 return IPPROTO_DONE;
223 }
224 #endif
225
226 void
227 tcp_input(struct mbuf *m, int off0)
228 {
229 struct tcphdr *th;
230 struct ip *ip = NULL;
231 struct ipovly *ipov;
232 struct inpcb *inp = NULL;
233 struct tcpcb *tp = NULL;
234 struct socket *so = NULL;
235 u_char *optp = NULL;
236 int optlen = 0;
237 int len, tlen, off;
238 int drop_hdrlen;
239 int thflags;
240 int rstreason = 0; /* For badport_bandlim accounting purposes */
241 #ifdef IPFIREWALL_FORWARD
242 struct m_tag *fwd_tag;
243 #endif
244 #ifdef INET6
245 struct ip6_hdr *ip6 = NULL;
246 int isipv6;
247 #else
248 const void *ip6 = NULL;
249 const int isipv6 = 0;
250 #endif
251 struct tcpopt to; /* options in this segment */
252 char *s = NULL; /* address and port logging */
253
254 #ifdef TCPDEBUG
255 /*
256 * The size of tcp_saveipgen must be the size of the max ip header,
257 * now IPv6.
258 */
259 u_char tcp_saveipgen[IP6_HDR_LEN];
260 struct tcphdr tcp_savetcp;
261 short ostate = 0;
262 #endif
263
264 #ifdef INET6
265 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
266 #endif
267
268 to.to_flags = 0;
269 tcpstat.tcps_rcvtotal++;
270
271 if (isipv6) {
272 #ifdef INET6
273 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
274 ip6 = mtod(m, struct ip6_hdr *);
275 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
276 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
277 tcpstat.tcps_rcvbadsum++;
278 goto drop;
279 }
280 th = (struct tcphdr *)((caddr_t)ip6 + off0);
281
282 /*
283 * Be proactive about unspecified IPv6 address in source.
284 * As we use all-zero to indicate unbounded/unconnected pcb,
285 * unspecified IPv6 address can be used to confuse us.
286 *
287 * Note that packets with unspecified IPv6 destination is
288 * already dropped in ip6_input.
289 */
290 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
291 /* XXX stat */
292 goto drop;
293 }
294 #else
295 th = NULL; /* XXX: Avoid compiler warning. */
296 #endif
297 } else {
298 /*
299 * Get IP and TCP header together in first mbuf.
300 * Note: IP leaves IP header in first mbuf.
301 */
302 if (off0 > sizeof (struct ip)) {
303 ip_stripoptions(m, (struct mbuf *)0);
304 off0 = sizeof(struct ip);
305 }
306 if (m->m_len < sizeof (struct tcpiphdr)) {
307 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
308 == NULL) {
309 tcpstat.tcps_rcvshort++;
310 return;
311 }
312 }
313 ip = mtod(m, struct ip *);
314 ipov = (struct ipovly *)ip;
315 th = (struct tcphdr *)((caddr_t)ip + off0);
316 tlen = ip->ip_len;
317
318 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
319 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
320 th->th_sum = m->m_pkthdr.csum_data;
321 else
322 th->th_sum = in_pseudo(ip->ip_src.s_addr,
323 ip->ip_dst.s_addr,
324 htonl(m->m_pkthdr.csum_data +
325 ip->ip_len +
326 IPPROTO_TCP));
327 th->th_sum ^= 0xffff;
328 #ifdef TCPDEBUG
329 ipov->ih_len = (u_short)tlen;
330 ipov->ih_len = htons(ipov->ih_len);
331 #endif
332 } else {
333 /*
334 * Checksum extended TCP header and data.
335 */
336 len = sizeof (struct ip) + tlen;
337 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
338 ipov->ih_len = (u_short)tlen;
339 ipov->ih_len = htons(ipov->ih_len);
340 th->th_sum = in_cksum(m, len);
341 }
342 if (th->th_sum) {
343 tcpstat.tcps_rcvbadsum++;
344 goto drop;
345 }
346 /* Re-initialization for later version check */
347 ip->ip_v = IPVERSION;
348 }
349
350 /*
351 * Check that TCP offset makes sense,
352 * pull out TCP options and adjust length. XXX
353 */
354 off = th->th_off << 2;
355 if (off < sizeof (struct tcphdr) || off > tlen) {
356 tcpstat.tcps_rcvbadoff++;
357 goto drop;
358 }
359 tlen -= off; /* tlen is used instead of ti->ti_len */
360 if (off > sizeof (struct tcphdr)) {
361 if (isipv6) {
362 #ifdef INET6
363 IP6_EXTHDR_CHECK(m, off0, off, );
364 ip6 = mtod(m, struct ip6_hdr *);
365 th = (struct tcphdr *)((caddr_t)ip6 + off0);
366 #endif
367 } else {
368 if (m->m_len < sizeof(struct ip) + off) {
369 if ((m = m_pullup(m, sizeof (struct ip) + off))
370 == NULL) {
371 tcpstat.tcps_rcvshort++;
372 return;
373 }
374 ip = mtod(m, struct ip *);
375 ipov = (struct ipovly *)ip;
376 th = (struct tcphdr *)((caddr_t)ip + off0);
377 }
378 }
379 optlen = off - sizeof (struct tcphdr);
380 optp = (u_char *)(th + 1);
381 }
382 thflags = th->th_flags;
383
384 /*
385 * Convert TCP protocol specific fields to host format.
386 */
387 th->th_seq = ntohl(th->th_seq);
388 th->th_ack = ntohl(th->th_ack);
389 th->th_win = ntohs(th->th_win);
390 th->th_urp = ntohs(th->th_urp);
391
392 /*
393 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
394 */
395 drop_hdrlen = off0 + off;
396
397 /*
398 * Locate pcb for segment.
399 */
400 INP_INFO_WLOCK(&tcbinfo);
401 findpcb:
402 INP_INFO_WLOCK_ASSERT(&tcbinfo);
403 #ifdef IPFIREWALL_FORWARD
404 /*
405 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
406 */
407 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
408
409 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */
410 struct sockaddr_in *next_hop;
411
412 next_hop = (struct sockaddr_in *)(fwd_tag+1);
413 /*
414 * Transparently forwarded. Pretend to be the destination.
415 * already got one like this?
416 */
417 inp = in_pcblookup_hash(&tcbinfo,
418 ip->ip_src, th->th_sport,
419 ip->ip_dst, th->th_dport,
420 0, m->m_pkthdr.rcvif);
421 if (!inp) {
422 /* It's new. Try to find the ambushing socket. */
423 inp = in_pcblookup_hash(&tcbinfo,
424 ip->ip_src, th->th_sport,
425 next_hop->sin_addr,
426 next_hop->sin_port ?
427 ntohs(next_hop->sin_port) :
428 th->th_dport,
429 INPLOOKUP_WILDCARD,
430 m->m_pkthdr.rcvif);
431 }
432 /* Remove the tag from the packet. We don't need it anymore. */
433 m_tag_delete(m, fwd_tag);
434 } else
435 #endif /* IPFIREWALL_FORWARD */
436 {
437 if (isipv6) {
438 #ifdef INET6
439 inp = in6_pcblookup_hash(&tcbinfo,
440 &ip6->ip6_src, th->th_sport,
441 &ip6->ip6_dst, th->th_dport,
442 INPLOOKUP_WILDCARD,
443 m->m_pkthdr.rcvif);
444 #endif
445 } else
446 inp = in_pcblookup_hash(&tcbinfo,
447 ip->ip_src, th->th_sport,
448 ip->ip_dst, th->th_dport,
449 INPLOOKUP_WILDCARD,
450 m->m_pkthdr.rcvif);
451 }
452
453 /*
454 * If the INPCB does not exist then all data in the incoming
455 * segment is discarded and an appropriate RST is sent back.
456 * XXX MRT Send RST using which routing table?
457 */
458 if (inp == NULL) {
459 /*
460 * Log communication attempts to ports that are not
461 * in use.
462 */
463 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
464 tcp_log_in_vain == 2) {
465 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
466 log(LOG_INFO, "%s; %s: Connection attempt "
467 "to closed port\n", s, __func__);
468 }
469 /*
470 * When blackholing do not respond with a RST but
471 * completely ignore the segment and drop it.
472 */
473 if ((blackhole == 1 && (thflags & TH_SYN)) ||
474 blackhole == 2)
475 goto dropunlock;
476
477 rstreason = BANDLIM_RST_CLOSEDPORT;
478 goto dropwithreset;
479 }
480 INP_WLOCK(inp);
481
482 #ifdef IPSEC
483 #ifdef INET6
484 if (isipv6 && ipsec6_in_reject(m, inp)) {
485 ipsec6stat.in_polvio++;
486 goto dropunlock;
487 } else
488 #endif /* INET6 */
489 if (ipsec4_in_reject(m, inp) != 0) {
490 ipsec4stat.in_polvio++;
491 goto dropunlock;
492 }
493 #endif /* IPSEC */
494
495 /*
496 * Check the minimum TTL for socket.
497 */
498 if (inp->inp_ip_minttl != 0) {
499 #ifdef INET6
500 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
501 goto dropunlock;
502 else
503 #endif
504 if (inp->inp_ip_minttl > ip->ip_ttl)
505 goto dropunlock;
506 }
507
508 /*
509 * A previous connection in TIMEWAIT state is supposed to catch
510 * stray or duplicate segments arriving late. If this segment
511 * was a legitimate new connection attempt the old INPCB gets
512 * removed and we can try again to find a listening socket.
513 */
514 if (inp->inp_flags & INP_TIMEWAIT) {
515 if (thflags & TH_SYN)
516 tcp_dooptions(&to, optp, optlen, TO_SYN);
517 /*
518 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
519 */
520 if (tcp_twcheck(inp, &to, th, m, tlen))
521 goto findpcb;
522 INP_INFO_WUNLOCK(&tcbinfo);
523 return;
524 }
525 /*
526 * The TCPCB may no longer exist if the connection is winding
527 * down or it is in the CLOSED state. Either way we drop the
528 * segment and send an appropriate response.
529 */
530 tp = intotcpcb(inp);
531 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
532 rstreason = BANDLIM_RST_CLOSEDPORT;
533 goto dropwithreset;
534 }
535
536 #ifdef MAC
537 INP_WLOCK_ASSERT(inp);
538 if (mac_check_inpcb_deliver(inp, m))
539 goto dropunlock;
540 #endif
541 so = inp->inp_socket;
542 KASSERT(so != NULL, ("%s: so == NULL", __func__));
543 #ifdef TCPDEBUG
544 if (so->so_options & SO_DEBUG) {
545 ostate = tp->t_state;
546 if (isipv6) {
547 #ifdef INET6
548 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
549 #endif
550 } else
551 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
552 tcp_savetcp = *th;
553 }
554 #endif
555 /*
556 * When the socket is accepting connections (the INPCB is in LISTEN
557 * state) we look into the SYN cache if this is a new connection
558 * attempt or the completion of a previous one.
559 */
560 if (so->so_options & SO_ACCEPTCONN) {
561 struct in_conninfo inc;
562
563 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
564 "tp not listening", __func__));
565
566 bzero(&inc, sizeof(inc));
567 #ifdef INET6
568 if (isipv6) {
569 inc.inc_flags |= INC_ISIPV6;
570 inc.inc6_faddr = ip6->ip6_src;
571 inc.inc6_laddr = ip6->ip6_dst;
572 } else
573 #endif
574 {
575 inc.inc_faddr = ip->ip_src;
576 inc.inc_laddr = ip->ip_dst;
577 }
578 inc.inc_fport = th->th_sport;
579 inc.inc_lport = th->th_dport;
580 inc.inc_fibnum = so->so_fibnum;
581
582 /*
583 * Check for an existing connection attempt in syncache if
584 * the flag is only ACK. A successful lookup creates a new
585 * socket appended to the listen queue in SYN_RECEIVED state.
586 */
587 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
588 /*
589 * Parse the TCP options here because
590 * syncookies need access to the reflected
591 * timestamp.
592 */
593 tcp_dooptions(&to, optp, optlen, 0);
594 /*
595 * NB: syncache_expand() doesn't unlock
596 * inp and tcpinfo locks.
597 */
598 if (!syncache_expand(&inc, &to, th, &so, m)) {
599 /*
600 * No syncache entry or ACK was not
601 * for our SYN/ACK. Send a RST.
602 * NB: syncache did its own logging
603 * of the failure cause.
604 */
605 rstreason = BANDLIM_RST_OPENPORT;
606 goto dropwithreset;
607 }
608 if (so == NULL) {
609 /*
610 * We completed the 3-way handshake
611 * but could not allocate a socket
612 * either due to memory shortage,
613 * listen queue length limits or
614 * global socket limits. Send RST
615 * or wait and have the remote end
616 * retransmit the ACK for another
617 * try.
618 */
619 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
620 log(LOG_DEBUG, "%s; %s: Listen socket: "
621 "Socket allocation failed due to "
622 "limits or memory shortage, %s\n",
623 s, __func__, (tcp_sc_rst_sock_fail ?
624 "sending RST" : "try again"));
625 if (tcp_sc_rst_sock_fail) {
626 rstreason = BANDLIM_UNLIMITED;
627 goto dropwithreset;
628 } else
629 goto dropunlock;
630 }
631 /*
632 * Socket is created in state SYN_RECEIVED.
633 * Unlock the listen socket, lock the newly
634 * created socket and update the tp variable.
635 */
636 INP_WUNLOCK(inp); /* listen socket */
637 inp = sotoinpcb(so);
638 INP_WLOCK(inp); /* new connection */
639 tp = intotcpcb(inp);
640 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
641 ("%s: ", __func__));
642 /*
643 * Process the segment and the data it
644 * contains. tcp_do_segment() consumes
645 * the mbuf chain and unlocks the inpcb.
646 */
647 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen);
648 INP_INFO_UNLOCK_ASSERT(&tcbinfo);
649 return;
650 }
651 /*
652 * Segment flag validation for new connection attempts:
653 *
654 * Our (SYN|ACK) response was rejected.
655 * Check with syncache and remove entry to prevent
656 * retransmits.
657 *
658 * NB: syncache_chkrst does its own logging of failure
659 * causes.
660 */
661 if (thflags & TH_RST) {
662 syncache_chkrst(&inc, th);
663 goto dropunlock;
664 }
665 /*
666 * We can't do anything without SYN.
667 */
668 if ((thflags & TH_SYN) == 0) {
669 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
670 log(LOG_DEBUG, "%s; %s: Listen socket: "
671 "SYN is missing, segment ignored\n",
672 s, __func__);
673 tcpstat.tcps_badsyn++;
674 goto dropunlock;
675 }
676 /*
677 * (SYN|ACK) is bogus on a listen socket.
678 */
679 if (thflags & TH_ACK) {
680 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
681 log(LOG_DEBUG, "%s; %s: Listen socket: "
682 "SYN|ACK invalid, segment rejected\n",
683 s, __func__);
684 syncache_badack(&inc); /* XXX: Not needed! */
685 tcpstat.tcps_badsyn++;
686 rstreason = BANDLIM_RST_OPENPORT;
687 goto dropwithreset;
688 }
689 /*
690 * If the drop_synfin option is enabled, drop all
691 * segments with both the SYN and FIN bits set.
692 * This prevents e.g. nmap from identifying the
693 * TCP/IP stack.
694 * XXX: Poor reasoning. nmap has other methods
695 * and is constantly refining its stack detection
696 * strategies.
697 * XXX: This is a violation of the TCP specification
698 * and was used by RFC1644.
699 */
700 if ((thflags & TH_FIN) && drop_synfin) {
701 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
702 log(LOG_DEBUG, "%s; %s: Listen socket: "
703 "SYN|FIN segment ignored (based on "
704 "sysctl setting)\n", s, __func__);
705 tcpstat.tcps_badsyn++;
706 goto dropunlock;
707 }
708 /*
709 * Segment's flags are (SYN) or (SYN|FIN).
710 *
711 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
712 * as they do not affect the state of the TCP FSM.
713 * The data pointed to by TH_URG and th_urp is ignored.
714 */
715 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
716 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
717 KASSERT(thflags & (TH_SYN),
718 ("%s: Listen socket: TH_SYN not set", __func__));
719 #ifdef INET6
720 /*
721 * If deprecated address is forbidden,
722 * we do not accept SYN to deprecated interface
723 * address to prevent any new inbound connection from
724 * getting established.
725 * When we do not accept SYN, we send a TCP RST,
726 * with deprecated source address (instead of dropping
727 * it). We compromise it as it is much better for peer
728 * to send a RST, and RST will be the final packet
729 * for the exchange.
730 *
731 * If we do not forbid deprecated addresses, we accept
732 * the SYN packet. RFC2462 does not suggest dropping
733 * SYN in this case.
734 * If we decipher RFC2462 5.5.4, it says like this:
735 * 1. use of deprecated addr with existing
736 * communication is okay - "SHOULD continue to be
737 * used"
738 * 2. use of it with new communication:
739 * (2a) "SHOULD NOT be used if alternate address
740 * with sufficient scope is available"
741 * (2b) nothing mentioned otherwise.
742 * Here we fall into (2b) case as we have no choice in
743 * our source address selection - we must obey the peer.
744 *
745 * The wording in RFC2462 is confusing, and there are
746 * multiple description text for deprecated address
747 * handling - worse, they are not exactly the same.
748 * I believe 5.5.4 is the best one, so we follow 5.5.4.
749 */
750 if (isipv6 && !ip6_use_deprecated) {
751 struct in6_ifaddr *ia6;
752
753 if ((ia6 = ip6_getdstifaddr(m)) &&
754 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
755 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
756 log(LOG_DEBUG, "%s; %s: Listen socket: "
757 "Connection attempt to deprecated "
758 "IPv6 address rejected\n",
759 s, __func__);
760 rstreason = BANDLIM_RST_OPENPORT;
761 goto dropwithreset;
762 }
763 }
764 #endif
765 /*
766 * Basic sanity checks on incoming SYN requests:
767 * Don't respond if the destination is a link layer
768 * broadcast according to RFC1122 4.2.3.10, p. 104.
769 * If it is from this socket it must be forged.
770 * Don't respond if the source or destination is a
771 * global or subnet broad- or multicast address.
772 * Note that it is quite possible to receive unicast
773 * link-layer packets with a broadcast IP address. Use
774 * in_broadcast() to find them.
775 */
776 if (m->m_flags & (M_BCAST|M_MCAST)) {
777 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
778 log(LOG_DEBUG, "%s; %s: Listen socket: "
779 "Connection attempt from broad- or multicast "
780 "link layer address ignored\n", s, __func__);
781 goto dropunlock;
782 }
783 if (isipv6) {
784 #ifdef INET6
785 if (th->th_dport == th->th_sport &&
786 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
787 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
788 log(LOG_DEBUG, "%s; %s: Listen socket: "
789 "Connection attempt to/from self "
790 "ignored\n", s, __func__);
791 goto dropunlock;
792 }
793 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
794 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
795 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
796 log(LOG_DEBUG, "%s; %s: Listen socket: "
797 "Connection attempt from/to multicast "
798 "address ignored\n", s, __func__);
799 goto dropunlock;
800 }
801 #endif
802 } else {
803 if (th->th_dport == th->th_sport &&
804 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
805 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
806 log(LOG_DEBUG, "%s; %s: Listen socket: "
807 "Connection attempt from/to self "
808 "ignored\n", s, __func__);
809 goto dropunlock;
810 }
811 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
812 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
813 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
814 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
815 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
816 log(LOG_DEBUG, "%s; %s: Listen socket: "
817 "Connection attempt from/to broad- "
818 "or multicast address ignored\n",
819 s, __func__);
820 goto dropunlock;
821 }
822 }
823 /*
824 * SYN appears to be valid. Create compressed TCP state
825 * for syncache.
826 */
827 #ifdef TCPDEBUG
828 if (so->so_options & SO_DEBUG)
829 tcp_trace(TA_INPUT, ostate, tp,
830 (void *)tcp_saveipgen, &tcp_savetcp, 0);
831 #endif
832 tcp_dooptions(&to, optp, optlen, TO_SYN);
833 syncache_add(&inc, &to, th, inp, &so, m);
834 /*
835 * Entry added to syncache and mbuf consumed.
836 * Everything already unlocked by syncache_add().
837 */
838 INP_INFO_UNLOCK_ASSERT(&tcbinfo);
839 return;
840 }
841
842 /*
843 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
844 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
845 * the inpcb, and unlocks pcbinfo.
846 */
847 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen);
848 INP_INFO_UNLOCK_ASSERT(&tcbinfo);
849 return;
850
851 dropwithreset:
852 INP_INFO_WLOCK_ASSERT(&tcbinfo);
853 INP_INFO_WUNLOCK(&tcbinfo);
854
855 if (inp != NULL) {
856 tcp_dropwithreset(m, th, tp, tlen, rstreason);
857 INP_WUNLOCK(inp);
858 } else
859 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
860 m = NULL; /* mbuf chain got consumed. */
861 goto drop;
862
863 dropunlock:
864 INP_INFO_WLOCK_ASSERT(&tcbinfo);
865 if (inp != NULL)
866 INP_WUNLOCK(inp);
867 INP_INFO_WUNLOCK(&tcbinfo);
868
869 drop:
870 INP_INFO_UNLOCK_ASSERT(&tcbinfo);
871 if (s != NULL)
872 free(s, M_TCPLOG);
873 if (m != NULL)
874 m_freem(m);
875 }
876
877 static void
878 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
879 struct tcpcb *tp, int drop_hdrlen, int tlen)
880 {
881 int thflags, acked, ourfinisacked, needoutput = 0;
882 int headlocked = 1;
883 int rstreason, todrop, win;
884 u_long tiwin;
885 struct tcpopt to;
886
887 #ifdef TCPDEBUG
888 /*
889 * The size of tcp_saveipgen must be the size of the max ip header,
890 * now IPv6.
891 */
892 u_char tcp_saveipgen[IP6_HDR_LEN];
893 struct tcphdr tcp_savetcp;
894 short ostate = 0;
895 #endif
896 thflags = th->th_flags;
897
898 INP_INFO_WLOCK_ASSERT(&tcbinfo);
899 INP_WLOCK_ASSERT(tp->t_inpcb);
900 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
901 __func__));
902 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
903 __func__));
904
905 /*
906 * Segment received on connection.
907 * Reset idle time and keep-alive timer.
908 * XXX: This should be done after segment
909 * validation to ignore broken/spoofed segs.
910 */
911 tp->t_rcvtime = ticks;
912 if (TCPS_HAVEESTABLISHED(tp->t_state))
913 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
914
915 /*
916 * Unscale the window into a 32-bit value.
917 * For the SYN_SENT state the scale is zero.
918 */
919 tiwin = th->th_win << tp->snd_scale;
920
921 /*
922 * Parse options on any incoming segment.
923 */
924 tcp_dooptions(&to, (u_char *)(th + 1),
925 (th->th_off << 2) - sizeof(struct tcphdr),
926 (thflags & TH_SYN) ? TO_SYN : 0);
927
928 /*
929 * If echoed timestamp is later than the current time,
930 * fall back to non RFC1323 RTT calculation. Normalize
931 * timestamp if syncookies were used when this connection
932 * was established.
933 */
934 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
935 to.to_tsecr -= tp->ts_offset;
936 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
937 to.to_tsecr = 0;
938 }
939
940 /*
941 * Process options only when we get SYN/ACK back. The SYN case
942 * for incoming connections is handled in tcp_syncache.
943 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
944 * or <SYN,ACK>) segment itself is never scaled.
945 * XXX this is traditional behavior, may need to be cleaned up.
946 */
947 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
948 if ((to.to_flags & TOF_SCALE) &&
949 (tp->t_flags & TF_REQ_SCALE)) {
950 tp->t_flags |= TF_RCVD_SCALE;
951 tp->snd_scale = to.to_wscale;
952 }
953 /*
954 * Initial send window. It will be updated with
955 * the next incoming segment to the scaled value.
956 */
957 tp->snd_wnd = th->th_win;
958 if (to.to_flags & TOF_TS) {
959 tp->t_flags |= TF_RCVD_TSTMP;
960 tp->ts_recent = to.to_tsval;
961 tp->ts_recent_age = tcp_ts_getticks();
962 }
963 if (to.to_flags & TOF_MSS)
964 tcp_mss(tp, to.to_mss);
965 if ((tp->t_flags & TF_SACK_PERMIT) &&
966 (to.to_flags & TOF_SACKPERM) == 0)
967 tp->t_flags &= ~TF_SACK_PERMIT;
968 }
969
970 /*
971 * Header prediction: check for the two common cases
972 * of a uni-directional data xfer. If the packet has
973 * no control flags, is in-sequence, the window didn't
974 * change and we're not retransmitting, it's a
975 * candidate. If the length is zero and the ack moved
976 * forward, we're the sender side of the xfer. Just
977 * free the data acked & wake any higher level process
978 * that was blocked waiting for space. If the length
979 * is non-zero and the ack didn't move, we're the
980 * receiver side. If we're getting packets in-order
981 * (the reassembly queue is empty), add the data to
982 * the socket buffer and note that we need a delayed ack.
983 * Make sure that the hidden state-flags are also off.
984 * Since we check for TCPS_ESTABLISHED first, it can only
985 * be TH_NEEDSYN.
986 */
987 if (tp->t_state == TCPS_ESTABLISHED &&
988 th->th_seq == tp->rcv_nxt &&
989 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
990 tp->snd_nxt == tp->snd_max &&
991 tiwin && tiwin == tp->snd_wnd &&
992 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
993 LIST_EMPTY(&tp->t_segq) &&
994 ((to.to_flags & TOF_TS) == 0 ||
995 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
996
997 /*
998 * If last ACK falls within this segment's sequence numbers,
999 * record the timestamp.
1000 * NOTE that the test is modified according to the latest
1001 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1002 */
1003 if ((to.to_flags & TOF_TS) != 0 &&
1004 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1005 tp->ts_recent_age = tcp_ts_getticks();
1006 tp->ts_recent = to.to_tsval;
1007 }
1008
1009 if (tlen == 0) {
1010 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1011 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1012 tp->snd_cwnd >= tp->snd_wnd &&
1013 ((!tcp_do_newreno &&
1014 !(tp->t_flags & TF_SACK_PERMIT) &&
1015 tp->t_dupacks < tcprexmtthresh) ||
1016 ((tcp_do_newreno ||
1017 (tp->t_flags & TF_SACK_PERMIT)) &&
1018 !IN_FASTRECOVERY(tp) &&
1019 (to.to_flags & TOF_SACK) == 0 &&
1020 TAILQ_EMPTY(&tp->snd_holes)))) {
1021 KASSERT(headlocked,
1022 ("%s: headlocked", __func__));
1023 INP_INFO_WUNLOCK(&tcbinfo);
1024 headlocked = 0;
1025 /*
1026 * This is a pure ack for outstanding data.
1027 */
1028 ++tcpstat.tcps_predack;
1029 /*
1030 * "bad retransmit" recovery.
1031 */
1032 if (tp->t_rxtshift == 1 &&
1033 tp->t_flags & TF_PREVVALID &&
1034 (int)(ticks - tp->t_badrxtwin) < 0) {
1035 ++tcpstat.tcps_sndrexmitbad;
1036 tp->snd_cwnd = tp->snd_cwnd_prev;
1037 tp->snd_ssthresh =
1038 tp->snd_ssthresh_prev;
1039 tp->snd_recover = tp->snd_recover_prev;
1040 if (tp->t_flags & TF_WASFRECOVERY)
1041 ENTER_FASTRECOVERY(tp);
1042 tp->snd_nxt = tp->snd_max;
1043 tp->t_flags &= ~TF_PREVVALID;
1044 tp->t_badrxtwin = 0;
1045 }
1046
1047 /*
1048 * Recalculate the transmit timer / rtt.
1049 *
1050 * Some boxes send broken timestamp replies
1051 * during the SYN+ACK phase, ignore
1052 * timestamps of 0 or we could calculate a
1053 * huge RTT and blow up the retransmit timer.
1054 */
1055 if ((to.to_flags & TOF_TS) != 0 &&
1056 to.to_tsecr) {
1057 u_int t;
1058
1059 t = tcp_ts_getticks() - to.to_tsecr;
1060 if (!tp->t_rttlow || tp->t_rttlow > t)
1061 tp->t_rttlow = t;
1062 tcp_xmit_timer(tp,
1063 TCP_TS_TO_TICKS(t) + 1);
1064 } else if (tp->t_rtttime &&
1065 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1066 if (!tp->t_rttlow ||
1067 tp->t_rttlow > ticks - tp->t_rtttime)
1068 tp->t_rttlow = ticks - tp->t_rtttime;
1069 tcp_xmit_timer(tp,
1070 ticks - tp->t_rtttime);
1071 }
1072 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1073 acked = th->th_ack - tp->snd_una;
1074 tcpstat.tcps_rcvackpack++;
1075 tcpstat.tcps_rcvackbyte += acked;
1076 sbdrop(&so->so_snd, acked);
1077 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1078 SEQ_LEQ(th->th_ack, tp->snd_recover))
1079 tp->snd_recover = th->th_ack - 1;
1080 tp->snd_una = th->th_ack;
1081 /*
1082 * Pull snd_wl2 up to prevent seq wrap relative
1083 * to th_ack.
1084 */
1085 tp->snd_wl2 = th->th_ack;
1086 tp->t_dupacks = 0;
1087 m_freem(m);
1088 ND6_HINT(tp); /* Some progress has been made. */
1089
1090 /*
1091 * If all outstanding data are acked, stop
1092 * retransmit timer, otherwise restart timer
1093 * using current (possibly backed-off) value.
1094 * If process is waiting for space,
1095 * wakeup/selwakeup/signal. If data
1096 * are ready to send, let tcp_output
1097 * decide between more output or persist.
1098 */
1099 #ifdef TCPDEBUG
1100 if (so->so_options & SO_DEBUG)
1101 tcp_trace(TA_INPUT, ostate, tp,
1102 (void *)tcp_saveipgen,
1103 &tcp_savetcp, 0);
1104 #endif
1105 if (tp->snd_una == tp->snd_max)
1106 tcp_timer_activate(tp, TT_REXMT, 0);
1107 else if (!tcp_timer_active(tp, TT_PERSIST))
1108 tcp_timer_activate(tp, TT_REXMT,
1109 tp->t_rxtcur);
1110 sowwakeup(so);
1111 if (so->so_snd.sb_cc)
1112 (void) tcp_output(tp);
1113 goto check_delack;
1114 }
1115 } else if (th->th_ack == tp->snd_una &&
1116 tlen <= sbspace(&so->so_rcv)) {
1117 int newsize = 0; /* automatic sockbuf scaling */
1118
1119 KASSERT(headlocked, ("%s: headlocked", __func__));
1120 INP_INFO_WUNLOCK(&tcbinfo);
1121 headlocked = 0;
1122 /*
1123 * This is a pure, in-sequence data packet
1124 * with nothing on the reassembly queue and
1125 * we have enough buffer space to take it.
1126 */
1127 /* Clean receiver SACK report if present */
1128 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1129 tcp_clean_sackreport(tp);
1130 ++tcpstat.tcps_preddat;
1131 tp->rcv_nxt += tlen;
1132 /*
1133 * Pull snd_wl1 up to prevent seq wrap relative to
1134 * th_seq.
1135 */
1136 tp->snd_wl1 = th->th_seq;
1137 /*
1138 * Pull rcv_up up to prevent seq wrap relative to
1139 * rcv_nxt.
1140 */
1141 tp->rcv_up = tp->rcv_nxt;
1142 tcpstat.tcps_rcvpack++;
1143 tcpstat.tcps_rcvbyte += tlen;
1144 ND6_HINT(tp); /* Some progress has been made */
1145 #ifdef TCPDEBUG
1146 if (so->so_options & SO_DEBUG)
1147 tcp_trace(TA_INPUT, ostate, tp,
1148 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1149 #endif
1150 /*
1151 * Automatic sizing of receive socket buffer. Often the send
1152 * buffer size is not optimally adjusted to the actual network
1153 * conditions at hand (delay bandwidth product). Setting the
1154 * buffer size too small limits throughput on links with high
1155 * bandwidth and high delay (eg. trans-continental/oceanic links).
1156 *
1157 * On the receive side the socket buffer memory is only rarely
1158 * used to any significant extent. This allows us to be much
1159 * more aggressive in scaling the receive socket buffer. For
1160 * the case that the buffer space is actually used to a large
1161 * extent and we run out of kernel memory we can simply drop
1162 * the new segments; TCP on the sender will just retransmit it
1163 * later. Setting the buffer size too big may only consume too
1164 * much kernel memory if the application doesn't read() from
1165 * the socket or packet loss or reordering makes use of the
1166 * reassembly queue.
1167 *
1168 * The criteria to step up the receive buffer one notch are:
1169 * 1. the number of bytes received during the time it takes
1170 * one timestamp to be reflected back to us (the RTT);
1171 * 2. received bytes per RTT is within seven eighth of the
1172 * current socket buffer size;
1173 * 3. receive buffer size has not hit maximal automatic size;
1174 *
1175 * This algorithm does one step per RTT at most and only if
1176 * we receive a bulk stream w/o packet losses or reorderings.
1177 * Shrinking the buffer during idle times is not necessary as
1178 * it doesn't consume any memory when idle.
1179 *
1180 * TODO: Only step up if the application is actually serving
1181 * the buffer to better manage the socket buffer resources.
1182 */
1183 if (tcp_do_autorcvbuf &&
1184 to.to_tsecr &&
1185 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1186 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1187 to.to_tsecr - tp->rfbuf_ts < hz) {
1188 if (tp->rfbuf_cnt >
1189 (so->so_rcv.sb_hiwat / 8 * 7) &&
1190 so->so_rcv.sb_hiwat <
1191 tcp_autorcvbuf_max) {
1192 newsize =
1193 min(so->so_rcv.sb_hiwat +
1194 tcp_autorcvbuf_inc,
1195 tcp_autorcvbuf_max);
1196 }
1197 /* Start over with next RTT. */
1198 tp->rfbuf_ts = 0;
1199 tp->rfbuf_cnt = 0;
1200 } else
1201 tp->rfbuf_cnt += tlen; /* add up */
1202 }
1203
1204 /* Add data to socket buffer. */
1205 SOCKBUF_LOCK(&so->so_rcv);
1206 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1207 m_freem(m);
1208 } else {
1209 /*
1210 * Set new socket buffer size.
1211 * Give up when limit is reached.
1212 */
1213 if (newsize)
1214 if (!sbreserve_locked(&so->so_rcv,
1215 newsize, so, NULL))
1216 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1217 m_adj(m, drop_hdrlen); /* delayed header drop */
1218 sbappendstream_locked(&so->so_rcv, m);
1219 }
1220 /* NB: sorwakeup_locked() does an implicit unlock. */
1221 sorwakeup_locked(so);
1222 if (DELAY_ACK(tp)) {
1223 tp->t_flags |= TF_DELACK;
1224 } else {
1225 tp->t_flags |= TF_ACKNOW;
1226 tcp_output(tp);
1227 }
1228 goto check_delack;
1229 }
1230 }
1231
1232 /*
1233 * Calculate amount of space in receive window,
1234 * and then do TCP input processing.
1235 * Receive window is amount of space in rcv queue,
1236 * but not less than advertised window.
1237 */
1238 win = sbspace(&so->so_rcv);
1239 if (win < 0)
1240 win = 0;
1241 KASSERT(SEQ_GEQ(tp->rcv_adv, tp->rcv_nxt),
1242 ("tcp_input negative window: tp %p rcv_nxt %u rcv_adv %u", tp,
1243 tp->rcv_adv, tp->rcv_nxt));
1244 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1245
1246 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1247 tp->rfbuf_ts = 0;
1248 tp->rfbuf_cnt = 0;
1249
1250 switch (tp->t_state) {
1251
1252 /*
1253 * If the state is SYN_RECEIVED:
1254 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1255 */
1256 case TCPS_SYN_RECEIVED:
1257 if ((thflags & TH_ACK) &&
1258 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1259 SEQ_GT(th->th_ack, tp->snd_max))) {
1260 rstreason = BANDLIM_RST_OPENPORT;
1261 goto dropwithreset;
1262 }
1263 break;
1264
1265 /*
1266 * If the state is SYN_SENT:
1267 * if seg contains an ACK, but not for our SYN, drop the input.
1268 * if seg contains a RST, then drop the connection.
1269 * if seg does not contain SYN, then drop it.
1270 * Otherwise this is an acceptable SYN segment
1271 * initialize tp->rcv_nxt and tp->irs
1272 * if seg contains ack then advance tp->snd_una
1273 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1274 * arrange for segment to be acked (eventually)
1275 * continue processing rest of data/controls, beginning with URG
1276 */
1277 case TCPS_SYN_SENT:
1278 if ((thflags & TH_ACK) &&
1279 (SEQ_LEQ(th->th_ack, tp->iss) ||
1280 SEQ_GT(th->th_ack, tp->snd_max))) {
1281 rstreason = BANDLIM_UNLIMITED;
1282 goto dropwithreset;
1283 }
1284 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST))
1285 tp = tcp_drop(tp, ECONNREFUSED);
1286 if (thflags & TH_RST)
1287 goto drop;
1288 if (!(thflags & TH_SYN))
1289 goto drop;
1290
1291 tp->irs = th->th_seq;
1292 tcp_rcvseqinit(tp);
1293 if (thflags & TH_ACK) {
1294 tcpstat.tcps_connects++;
1295 soisconnected(so);
1296 #ifdef MAC
1297 SOCK_LOCK(so);
1298 mac_set_socket_peer_from_mbuf(m, so);
1299 SOCK_UNLOCK(so);
1300 #endif
1301 /* Do window scaling on this connection? */
1302 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1303 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1304 tp->rcv_scale = tp->request_r_scale;
1305 }
1306 tp->rcv_adv += imin(tp->rcv_wnd,
1307 TCP_MAXWIN << tp->rcv_scale);
1308 tp->snd_una++; /* SYN is acked */
1309 /*
1310 * If there's data, delay ACK; if there's also a FIN
1311 * ACKNOW will be turned on later.
1312 */
1313 if (DELAY_ACK(tp) && tlen != 0)
1314 tcp_timer_activate(tp, TT_DELACK,
1315 tcp_delacktime);
1316 else
1317 tp->t_flags |= TF_ACKNOW;
1318 /*
1319 * Received <SYN,ACK> in SYN_SENT[*] state.
1320 * Transitions:
1321 * SYN_SENT --> ESTABLISHED
1322 * SYN_SENT* --> FIN_WAIT_1
1323 */
1324 tp->t_starttime = ticks;
1325 if (tp->t_flags & TF_NEEDFIN) {
1326 tp->t_state = TCPS_FIN_WAIT_1;
1327 tp->t_flags &= ~TF_NEEDFIN;
1328 thflags &= ~TH_SYN;
1329 } else {
1330 tp->t_state = TCPS_ESTABLISHED;
1331 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
1332 }
1333 } else {
1334 /*
1335 * Received initial SYN in SYN-SENT[*] state =>
1336 * simultaneous open. If segment contains CC option
1337 * and there is a cached CC, apply TAO test.
1338 * If it succeeds, connection is * half-synchronized.
1339 * Otherwise, do 3-way handshake:
1340 * SYN-SENT -> SYN-RECEIVED
1341 * SYN-SENT* -> SYN-RECEIVED*
1342 * If there was no CC option, clear cached CC value.
1343 */
1344 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1345 tcp_timer_activate(tp, TT_REXMT, 0);
1346 tp->t_state = TCPS_SYN_RECEIVED;
1347 }
1348
1349 KASSERT(headlocked, ("%s: trimthenstep6: head not locked",
1350 __func__));
1351 INP_WLOCK_ASSERT(tp->t_inpcb);
1352
1353 /*
1354 * Advance th->th_seq to correspond to first data byte.
1355 * If data, trim to stay within window,
1356 * dropping FIN if necessary.
1357 */
1358 th->th_seq++;
1359 if (tlen > tp->rcv_wnd) {
1360 todrop = tlen - tp->rcv_wnd;
1361 m_adj(m, -todrop);
1362 tlen = tp->rcv_wnd;
1363 thflags &= ~TH_FIN;
1364 tcpstat.tcps_rcvpackafterwin++;
1365 tcpstat.tcps_rcvbyteafterwin += todrop;
1366 }
1367 tp->snd_wl1 = th->th_seq - 1;
1368 tp->rcv_up = th->th_seq;
1369 /*
1370 * Client side of transaction: already sent SYN and data.
1371 * If the remote host used T/TCP to validate the SYN,
1372 * our data will be ACK'd; if so, enter normal data segment
1373 * processing in the middle of step 5, ack processing.
1374 * Otherwise, goto step 6.
1375 */
1376 if (thflags & TH_ACK)
1377 goto process_ACK;
1378
1379 goto step6;
1380
1381 /*
1382 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1383 * do normal processing.
1384 *
1385 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
1386 */
1387 case TCPS_LAST_ACK:
1388 case TCPS_CLOSING:
1389 break; /* continue normal processing */
1390 }
1391
1392 /*
1393 * States other than LISTEN or SYN_SENT.
1394 * First check the RST flag and sequence number since reset segments
1395 * are exempt from the timestamp and connection count tests. This
1396 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1397 * below which allowed reset segments in half the sequence space
1398 * to fall though and be processed (which gives forged reset
1399 * segments with a random sequence number a 50 percent chance of
1400 * killing a connection).
1401 * Then check timestamp, if present.
1402 * Then check the connection count, if present.
1403 * Then check that at least some bytes of segment are within
1404 * receive window. If segment begins before rcv_nxt,
1405 * drop leading data (and SYN); if nothing left, just ack.
1406 *
1407 *
1408 * If the RST bit is set, check the sequence number to see
1409 * if this is a valid reset segment.
1410 * RFC 793 page 37:
1411 * In all states except SYN-SENT, all reset (RST) segments
1412 * are validated by checking their SEQ-fields. A reset is
1413 * valid if its sequence number is in the window.
1414 * Note: this does not take into account delayed ACKs, so
1415 * we should test against last_ack_sent instead of rcv_nxt.
1416 * The sequence number in the reset segment is normally an
1417 * echo of our outgoing acknowlegement numbers, but some hosts
1418 * send a reset with the sequence number at the rightmost edge
1419 * of our receive window, and we have to handle this case.
1420 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1421 * that brute force RST attacks are possible. To combat this,
1422 * we use a much stricter check while in the ESTABLISHED state,
1423 * only accepting RSTs where the sequence number is equal to
1424 * last_ack_sent. In all other states (the states in which a
1425 * RST is more likely), the more permissive check is used.
1426 * If we have multiple segments in flight, the initial reset
1427 * segment sequence numbers will be to the left of last_ack_sent,
1428 * but they will eventually catch up.
1429 * In any case, it never made sense to trim reset segments to
1430 * fit the receive window since RFC 1122 says:
1431 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1432 *
1433 * A TCP SHOULD allow a received RST segment to include data.
1434 *
1435 * DISCUSSION
1436 * It has been suggested that a RST segment could contain
1437 * ASCII text that encoded and explained the cause of the
1438 * RST. No standard has yet been established for such
1439 * data.
1440 *
1441 * If the reset segment passes the sequence number test examine
1442 * the state:
1443 * SYN_RECEIVED STATE:
1444 * If passive open, return to LISTEN state.
1445 * If active open, inform user that connection was refused.
1446 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1447 * Inform user that connection was reset, and close tcb.
1448 * CLOSING, LAST_ACK STATES:
1449 * Close the tcb.
1450 * TIME_WAIT STATE:
1451 * Drop the segment - see Stevens, vol. 2, p. 964 and
1452 * RFC 1337.
1453 */
1454 if (thflags & TH_RST) {
1455 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
1456 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1457 switch (tp->t_state) {
1458
1459 case TCPS_SYN_RECEIVED:
1460 so->so_error = ECONNREFUSED;
1461 goto close;
1462
1463 case TCPS_ESTABLISHED:
1464 if (tcp_insecure_rst == 0 &&
1465 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
1466 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
1467 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
1468 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
1469 tcpstat.tcps_badrst++;
1470 goto drop;
1471 }
1472 /* FALLTHROUGH */
1473 case TCPS_FIN_WAIT_1:
1474 case TCPS_FIN_WAIT_2:
1475 case TCPS_CLOSE_WAIT:
1476 so->so_error = ECONNRESET;
1477 close:
1478 tp->t_state = TCPS_CLOSED;
1479 tcpstat.tcps_drops++;
1480 KASSERT(headlocked, ("%s: trimthenstep6: "
1481 "tcp_close: head not locked", __func__));
1482 tp = tcp_close(tp);
1483 break;
1484
1485 case TCPS_CLOSING:
1486 case TCPS_LAST_ACK:
1487 KASSERT(headlocked, ("%s: trimthenstep6: "
1488 "tcp_close.2: head not locked", __func__));
1489 tp = tcp_close(tp);
1490 break;
1491 }
1492 }
1493 goto drop;
1494 }
1495
1496 /*
1497 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1498 * and it's less than ts_recent, drop it.
1499 */
1500 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1501 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1502
1503 /* Check to see if ts_recent is over 24 days old. */
1504 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
1505 /*
1506 * Invalidate ts_recent. If this segment updates
1507 * ts_recent, the age will be reset later and ts_recent
1508 * will get a valid value. If it does not, setting
1509 * ts_recent to zero will at least satisfy the
1510 * requirement that zero be placed in the timestamp
1511 * echo reply when ts_recent isn't valid. The
1512 * age isn't reset until we get a valid ts_recent
1513 * because we don't want out-of-order segments to be
1514 * dropped when ts_recent is old.
1515 */
1516 tp->ts_recent = 0;
1517 } else {
1518 tcpstat.tcps_rcvduppack++;
1519 tcpstat.tcps_rcvdupbyte += tlen;
1520 tcpstat.tcps_pawsdrop++;
1521 if (tlen)
1522 goto dropafterack;
1523 goto drop;
1524 }
1525 }
1526
1527 /*
1528 * In the SYN-RECEIVED state, validate that the packet belongs to
1529 * this connection before trimming the data to fit the receive
1530 * window. Check the sequence number versus IRS since we know
1531 * the sequence numbers haven't wrapped. This is a partial fix
1532 * for the "LAND" DoS attack.
1533 */
1534 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1535 rstreason = BANDLIM_RST_OPENPORT;
1536 goto dropwithreset;
1537 }
1538
1539 todrop = tp->rcv_nxt - th->th_seq;
1540 if (todrop > 0) {
1541 if (thflags & TH_SYN) {
1542 thflags &= ~TH_SYN;
1543 th->th_seq++;
1544 if (th->th_urp > 1)
1545 th->th_urp--;
1546 else
1547 thflags &= ~TH_URG;
1548 todrop--;
1549 }
1550 /*
1551 * Following if statement from Stevens, vol. 2, p. 960.
1552 */
1553 if (todrop > tlen
1554 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1555 /*
1556 * Any valid FIN must be to the left of the window.
1557 * At this point the FIN must be a duplicate or out
1558 * of sequence; drop it.
1559 */
1560 thflags &= ~TH_FIN;
1561
1562 /*
1563 * Send an ACK to resynchronize and drop any data.
1564 * But keep on processing for RST or ACK.
1565 */
1566 tp->t_flags |= TF_ACKNOW;
1567 todrop = tlen;
1568 tcpstat.tcps_rcvduppack++;
1569 tcpstat.tcps_rcvdupbyte += todrop;
1570 } else {
1571 tcpstat.tcps_rcvpartduppack++;
1572 tcpstat.tcps_rcvpartdupbyte += todrop;
1573 }
1574 drop_hdrlen += todrop; /* drop from the top afterwards */
1575 th->th_seq += todrop;
1576 tlen -= todrop;
1577 if (th->th_urp > todrop)
1578 th->th_urp -= todrop;
1579 else {
1580 thflags &= ~TH_URG;
1581 th->th_urp = 0;
1582 }
1583 }
1584
1585 /*
1586 * If new data are received on a connection after the
1587 * user processes are gone, then RST the other end.
1588 */
1589 if ((so->so_state & SS_NOFDREF) &&
1590 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1591 char *s;
1592
1593 KASSERT(headlocked, ("%s: trimthenstep6: tcp_close.3: head "
1594 "not locked", __func__));
1595 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) {
1596 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket "
1597 "was closed, sending RST and removing tcpcb\n",
1598 s, __func__, tcpstates[tp->t_state], tlen);
1599 free(s, M_TCPLOG);
1600 }
1601 tp = tcp_close(tp);
1602 tcpstat.tcps_rcvafterclose++;
1603 rstreason = BANDLIM_UNLIMITED;
1604 goto dropwithreset;
1605 }
1606
1607 /*
1608 * If segment ends after window, drop trailing data
1609 * (and PUSH and FIN); if nothing left, just ACK.
1610 */
1611 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1612 if (todrop > 0) {
1613 tcpstat.tcps_rcvpackafterwin++;
1614 if (todrop >= tlen) {
1615 tcpstat.tcps_rcvbyteafterwin += tlen;
1616 /*
1617 * If window is closed can only take segments at
1618 * window edge, and have to drop data and PUSH from
1619 * incoming segments. Continue processing, but
1620 * remember to ack. Otherwise, drop segment
1621 * and ack.
1622 */
1623 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1624 tp->t_flags |= TF_ACKNOW;
1625 tcpstat.tcps_rcvwinprobe++;
1626 } else
1627 goto dropafterack;
1628 } else
1629 tcpstat.tcps_rcvbyteafterwin += todrop;
1630 m_adj(m, -todrop);
1631 tlen -= todrop;
1632 thflags &= ~(TH_PUSH|TH_FIN);
1633 }
1634
1635 /*
1636 * If last ACK falls within this segment's sequence numbers,
1637 * record its timestamp.
1638 * NOTE:
1639 * 1) That the test incorporates suggestions from the latest
1640 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1641 * 2) That updating only on newer timestamps interferes with
1642 * our earlier PAWS tests, so this check should be solely
1643 * predicated on the sequence space of this segment.
1644 * 3) That we modify the segment boundary check to be
1645 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
1646 * instead of RFC1323's
1647 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
1648 * This modified check allows us to overcome RFC1323's
1649 * limitations as described in Stevens TCP/IP Illustrated
1650 * Vol. 2 p.869. In such cases, we can still calculate the
1651 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1652 */
1653 if ((to.to_flags & TOF_TS) != 0 &&
1654 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1655 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1656 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
1657 tp->ts_recent_age = tcp_ts_getticks();
1658 tp->ts_recent = to.to_tsval;
1659 }
1660
1661 /*
1662 * If a SYN is in the window, then this is an
1663 * error and we send an RST and drop the connection.
1664 */
1665 if (thflags & TH_SYN) {
1666 KASSERT(headlocked, ("%s: tcp_drop: trimthenstep6: "
1667 "head not locked", __func__));
1668 tp = tcp_drop(tp, ECONNRESET);
1669 rstreason = BANDLIM_UNLIMITED;
1670 goto drop;
1671 }
1672
1673 /*
1674 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1675 * flag is on (half-synchronized state), then queue data for
1676 * later processing; else drop segment and return.
1677 */
1678 if ((thflags & TH_ACK) == 0) {
1679 if (tp->t_state == TCPS_SYN_RECEIVED ||
1680 (tp->t_flags & TF_NEEDSYN))
1681 goto step6;
1682 else if (tp->t_flags & TF_ACKNOW)
1683 goto dropafterack;
1684 else
1685 goto drop;
1686 }
1687
1688 /*
1689 * Ack processing.
1690 */
1691 switch (tp->t_state) {
1692
1693 /*
1694 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1695 * ESTABLISHED state and continue processing.
1696 * The ACK was checked above.
1697 */
1698 case TCPS_SYN_RECEIVED:
1699
1700 tcpstat.tcps_connects++;
1701 soisconnected(so);
1702 /* Do window scaling? */
1703 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1704 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1705 tp->rcv_scale = tp->request_r_scale;
1706 tp->snd_wnd = tiwin;
1707 }
1708 /*
1709 * Make transitions:
1710 * SYN-RECEIVED -> ESTABLISHED
1711 * SYN-RECEIVED* -> FIN-WAIT-1
1712 */
1713 tp->t_starttime = ticks;
1714 if (tp->t_flags & TF_NEEDFIN) {
1715 tp->t_state = TCPS_FIN_WAIT_1;
1716 tp->t_flags &= ~TF_NEEDFIN;
1717 } else {
1718 tp->t_state = TCPS_ESTABLISHED;
1719 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
1720 }
1721 /*
1722 * If segment contains data or ACK, will call tcp_reass()
1723 * later; if not, do so now to pass queued data to user.
1724 */
1725 if (tlen == 0 && (thflags & TH_FIN) == 0)
1726 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1727 (struct mbuf *)0);
1728 tp->snd_wl1 = th->th_seq - 1;
1729 /* FALLTHROUGH */
1730
1731 /*
1732 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1733 * ACKs. If the ack is in the range
1734 * tp->snd_una < th->th_ack <= tp->snd_max
1735 * then advance tp->snd_una to th->th_ack and drop
1736 * data from the retransmission queue. If this ACK reflects
1737 * more up to date window information we update our window information.
1738 */
1739 case TCPS_ESTABLISHED:
1740 case TCPS_FIN_WAIT_1:
1741 case TCPS_FIN_WAIT_2:
1742 case TCPS_CLOSE_WAIT:
1743 case TCPS_CLOSING:
1744 case TCPS_LAST_ACK:
1745 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1746 tcpstat.tcps_rcvacktoomuch++;
1747 goto dropafterack;
1748 }
1749 if ((tp->t_flags & TF_SACK_PERMIT) &&
1750 ((to.to_flags & TOF_SACK) ||
1751 !TAILQ_EMPTY(&tp->snd_holes)))
1752 tcp_sack_doack(tp, &to, th->th_ack);
1753 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1754 if (tlen == 0 && tiwin == tp->snd_wnd) {
1755 tcpstat.tcps_rcvdupack++;
1756 /*
1757 * If we have outstanding data (other than
1758 * a window probe), this is a completely
1759 * duplicate ack (ie, window info didn't
1760 * change), the ack is the biggest we've
1761 * seen and we've seen exactly our rexmt
1762 * threshhold of them, assume a packet
1763 * has been dropped and retransmit it.
1764 * Kludge snd_nxt & the congestion
1765 * window so we send only this one
1766 * packet.
1767 *
1768 * We know we're losing at the current
1769 * window size so do congestion avoidance
1770 * (set ssthresh to half the current window
1771 * and pull our congestion window back to
1772 * the new ssthresh).
1773 *
1774 * Dup acks mean that packets have left the
1775 * network (they're now cached at the receiver)
1776 * so bump cwnd by the amount in the receiver
1777 * to keep a constant cwnd packets in the
1778 * network.
1779 */
1780 if (!tcp_timer_active(tp, TT_REXMT) ||
1781 th->th_ack != tp->snd_una)
1782 tp->t_dupacks = 0;
1783 else if (++tp->t_dupacks > tcprexmtthresh ||
1784 ((tcp_do_newreno ||
1785 (tp->t_flags & TF_SACK_PERMIT)) &&
1786 IN_FASTRECOVERY(tp))) {
1787 if ((tp->t_flags & TF_SACK_PERMIT) &&
1788 IN_FASTRECOVERY(tp)) {
1789 int awnd;
1790
1791 /*
1792 * Compute the amount of data in flight first.
1793 * We can inject new data into the pipe iff
1794 * we have less than 1/2 the original window's
1795 * worth of data in flight.
1796 */
1797 awnd = (tp->snd_nxt - tp->snd_fack) +
1798 tp->sackhint.sack_bytes_rexmit;
1799 if (awnd < tp->snd_ssthresh) {
1800 tp->snd_cwnd += tp->t_maxseg;
1801 if (tp->snd_cwnd > tp->snd_ssthresh)
1802 tp->snd_cwnd = tp->snd_ssthresh;
1803 }
1804 } else
1805 tp->snd_cwnd += tp->t_maxseg;
1806 if ((thflags & TH_FIN) &&
1807 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
1808 /*
1809 * If its a fin we need to process
1810 * it to avoid a race where both
1811 * sides enter FIN-WAIT and send FIN|ACK
1812 * at the same time.
1813 */
1814 break;
1815 }
1816 (void) tcp_output(tp);
1817 goto drop;
1818 } else if (tp->t_dupacks == tcprexmtthresh) {
1819 tcp_seq onxt = tp->snd_nxt;
1820 u_int win;
1821
1822 /*
1823 * If we're doing sack, check to
1824 * see if we're already in sack
1825 * recovery. If we're not doing sack,
1826 * check to see if we're in newreno
1827 * recovery.
1828 */
1829 if (tp->t_flags & TF_SACK_PERMIT) {
1830 if (IN_FASTRECOVERY(tp)) {
1831 tp->t_dupacks = 0;
1832 break;
1833 }
1834 } else if (tcp_do_newreno) {
1835 if (SEQ_LEQ(th->th_ack,
1836 tp->snd_recover)) {
1837 tp->t_dupacks = 0;
1838 break;
1839 }
1840 }
1841 win = min(tp->snd_wnd, tp->snd_cwnd) /
1842 2 / tp->t_maxseg;
1843 if (win < 2)
1844 win = 2;
1845 tp->snd_ssthresh = win * tp->t_maxseg;
1846 ENTER_FASTRECOVERY(tp);
1847 tp->snd_recover = tp->snd_max;
1848 tcp_timer_activate(tp, TT_REXMT, 0);
1849 tp->t_rtttime = 0;
1850 if (tp->t_flags & TF_SACK_PERMIT) {
1851 tcpstat.tcps_sack_recovery_episode++;
1852 tp->sack_newdata = tp->snd_nxt;
1853 tp->snd_cwnd = tp->t_maxseg;
1854 (void) tcp_output(tp);
1855 goto drop;
1856 }
1857 tp->snd_nxt = th->th_ack;
1858 tp->snd_cwnd = tp->t_maxseg;
1859 if ((thflags & TH_FIN) &&
1860 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
1861 /*
1862 * If its a fin we need to process
1863 * it to avoid a race where both
1864 * sides enter FIN-WAIT and send FIN|ACK
1865 * at the same time.
1866 */
1867 break;
1868 }
1869 (void) tcp_output(tp);
1870 KASSERT(tp->snd_limited <= 2,
1871 ("%s: tp->snd_limited too big",
1872 __func__));
1873 tp->snd_cwnd = tp->snd_ssthresh +
1874 tp->t_maxseg *
1875 (tp->t_dupacks - tp->snd_limited);
1876 if (SEQ_GT(onxt, tp->snd_nxt))
1877 tp->snd_nxt = onxt;
1878 goto drop;
1879 } else if (tcp_do_rfc3042) {
1880 u_long oldcwnd = tp->snd_cwnd;
1881 tcp_seq oldsndmax = tp->snd_max;
1882 u_int sent;
1883
1884 KASSERT(tp->t_dupacks == 1 ||
1885 tp->t_dupacks == 2,
1886 ("%s: dupacks not 1 or 2",
1887 __func__));
1888 if (tp->t_dupacks == 1)
1889 tp->snd_limited = 0;
1890 tp->snd_cwnd =
1891 (tp->snd_nxt - tp->snd_una) +
1892 (tp->t_dupacks - tp->snd_limited) *
1893 tp->t_maxseg;
1894 if ((thflags & TH_FIN) &&
1895 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
1896 /*
1897 * If its a fin we need to process
1898 * it to avoid a race where both
1899 * sides enter FIN-WAIT and send FIN|ACK
1900 * at the same time.
1901 */
1902 break;
1903 }
1904 (void) tcp_output(tp);
1905 sent = tp->snd_max - oldsndmax;
1906 if (sent > tp->t_maxseg) {
1907 KASSERT((tp->t_dupacks == 2 &&
1908 tp->snd_limited == 0) ||
1909 (sent == tp->t_maxseg + 1 &&
1910 tp->t_flags & TF_SENTFIN),
1911 ("%s: sent too much",
1912 __func__));
1913 tp->snd_limited = 2;
1914 } else if (sent > 0)
1915 ++tp->snd_limited;
1916 tp->snd_cwnd = oldcwnd;
1917 goto drop;
1918 }
1919 } else
1920 tp->t_dupacks = 0;
1921 break;
1922 }
1923
1924 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
1925 ("%s: th_ack <= snd_una", __func__));
1926
1927 /*
1928 * If the congestion window was inflated to account
1929 * for the other side's cached packets, retract it.
1930 */
1931 if (tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) {
1932 if (IN_FASTRECOVERY(tp)) {
1933 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1934 if (tp->t_flags & TF_SACK_PERMIT)
1935 tcp_sack_partialack(tp, th);
1936 else
1937 tcp_newreno_partial_ack(tp, th);
1938 } else {
1939 /*
1940 * Out of fast recovery.
1941 * Window inflation should have left us
1942 * with approximately snd_ssthresh
1943 * outstanding data.
1944 * But in case we would be inclined to
1945 * send a burst, better to do it via
1946 * the slow start mechanism.
1947 */
1948 if (SEQ_GT(th->th_ack +
1949 tp->snd_ssthresh,
1950 tp->snd_max))
1951 tp->snd_cwnd = tp->snd_max -
1952 th->th_ack +
1953 tp->t_maxseg;
1954 else
1955 tp->snd_cwnd = tp->snd_ssthresh;
1956 }
1957 }
1958 } else {
1959 if (tp->t_dupacks >= tcprexmtthresh &&
1960 tp->snd_cwnd > tp->snd_ssthresh)
1961 tp->snd_cwnd = tp->snd_ssthresh;
1962 }
1963 tp->t_dupacks = 0;
1964 /*
1965 * If we reach this point, ACK is not a duplicate,
1966 * i.e., it ACKs something we sent.
1967 */
1968 if (tp->t_flags & TF_NEEDSYN) {
1969 /*
1970 * T/TCP: Connection was half-synchronized, and our
1971 * SYN has been ACK'd (so connection is now fully
1972 * synchronized). Go to non-starred state,
1973 * increment snd_una for ACK of SYN, and check if
1974 * we can do window scaling.
1975 */
1976 tp->t_flags &= ~TF_NEEDSYN;
1977 tp->snd_una++;
1978 /* Do window scaling? */
1979 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1980 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1981 tp->rcv_scale = tp->request_r_scale;
1982 /* Send window already scaled. */
1983 }
1984 }
1985
1986 process_ACK:
1987 KASSERT(headlocked, ("%s: process_ACK: head not locked",
1988 __func__));
1989 INP_WLOCK_ASSERT(tp->t_inpcb);
1990
1991 acked = th->th_ack - tp->snd_una;
1992 tcpstat.tcps_rcvackpack++;
1993 tcpstat.tcps_rcvackbyte += acked;
1994
1995 /*
1996 * If we just performed our first retransmit, and the ACK
1997 * arrives within our recovery window, then it was a mistake
1998 * to do the retransmit in the first place. Recover our
1999 * original cwnd and ssthresh, and proceed to transmit where
2000 * we left off.
2001 */
2002 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2003 (int)(ticks - tp->t_badrxtwin) < 0) {
2004 ++tcpstat.tcps_sndrexmitbad;
2005 tp->snd_cwnd = tp->snd_cwnd_prev;
2006 tp->snd_ssthresh = tp->snd_ssthresh_prev;
2007 tp->snd_recover = tp->snd_recover_prev;
2008 if (tp->t_flags & TF_WASFRECOVERY)
2009 ENTER_FASTRECOVERY(tp);
2010 tp->snd_nxt = tp->snd_max;
2011 tp->t_flags &= ~TF_PREVVALID;
2012 tp->t_badrxtwin = 0; /* XXX probably not required */
2013 }
2014
2015 /*
2016 * If we have a timestamp reply, update smoothed
2017 * round trip time. If no timestamp is present but
2018 * transmit timer is running and timed sequence
2019 * number was acked, update smoothed round trip time.
2020 * Since we now have an rtt measurement, cancel the
2021 * timer backoff (cf., Phil Karn's retransmit alg.).
2022 * Recompute the initial retransmit timer.
2023 *
2024 * Some boxes send broken timestamp replies
2025 * during the SYN+ACK phase, ignore
2026 * timestamps of 0 or we could calculate a
2027 * huge RTT and blow up the retransmit timer.
2028 */
2029 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2030 u_int t;
2031
2032 t = tcp_ts_getticks() - to.to_tsecr;
2033 if (!tp->t_rttlow || tp->t_rttlow > t)
2034 tp->t_rttlow = t;
2035 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2036 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2037 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2038 tp->t_rttlow = ticks - tp->t_rtttime;
2039 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2040 }
2041 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2042
2043 /*
2044 * If all outstanding data is acked, stop retransmit
2045 * timer and remember to restart (more output or persist).
2046 * If there is more data to be acked, restart retransmit
2047 * timer, using current (possibly backed-off) value.
2048 */
2049 if (th->th_ack == tp->snd_max) {
2050 tcp_timer_activate(tp, TT_REXMT, 0);
2051 needoutput = 1;
2052 } else if (!tcp_timer_active(tp, TT_PERSIST))
2053 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2054
2055 /*
2056 * If no data (only SYN) was ACK'd,
2057 * skip rest of ACK processing.
2058 */
2059 if (acked == 0)
2060 goto step6;
2061
2062 /*
2063 * When new data is acked, open the congestion window.
2064 * If the window gives us less than ssthresh packets
2065 * in flight, open exponentially (maxseg per packet).
2066 * Otherwise open linearly: maxseg per window
2067 * (maxseg^2 / cwnd per packet).
2068 * If cwnd > maxseg^2, fix the cwnd increment at 1 byte
2069 * to avoid capping cwnd (as suggested in RFC2581).
2070 */
2071 if ((!tcp_do_newreno && !(tp->t_flags & TF_SACK_PERMIT)) ||
2072 !IN_FASTRECOVERY(tp)) {
2073 u_int cw = tp->snd_cwnd;
2074 u_int incr = tp->t_maxseg;
2075 if (cw > tp->snd_ssthresh)
2076 incr = max((incr * incr / cw), 1);
2077 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2078 }
2079 SOCKBUF_LOCK(&so->so_snd);
2080 if (acked > so->so_snd.sb_cc) {
2081 tp->snd_wnd -= so->so_snd.sb_cc;
2082 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2083 ourfinisacked = 1;
2084 } else {
2085 sbdrop_locked(&so->so_snd, acked);
2086 tp->snd_wnd -= acked;
2087 ourfinisacked = 0;
2088 }
2089 /* NB: sowwakeup_locked() does an implicit unlock. */
2090 sowwakeup_locked(so);
2091 /* Detect una wraparound. */
2092 if ((tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) &&
2093 !IN_FASTRECOVERY(tp) &&
2094 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2095 SEQ_LEQ(th->th_ack, tp->snd_recover))
2096 tp->snd_recover = th->th_ack - 1;
2097 if ((tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) &&
2098 IN_FASTRECOVERY(tp) &&
2099 SEQ_GEQ(th->th_ack, tp->snd_recover))
2100 EXIT_FASTRECOVERY(tp);
2101 tp->snd_una = th->th_ack;
2102 if (tp->t_flags & TF_SACK_PERMIT) {
2103 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2104 tp->snd_recover = tp->snd_una;
2105 }
2106 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2107 tp->snd_nxt = tp->snd_una;
2108
2109 switch (tp->t_state) {
2110
2111 /*
2112 * In FIN_WAIT_1 STATE in addition to the processing
2113 * for the ESTABLISHED state if our FIN is now acknowledged
2114 * then enter FIN_WAIT_2.
2115 */
2116 case TCPS_FIN_WAIT_1:
2117 if (ourfinisacked) {
2118 /*
2119 * If we can't receive any more
2120 * data, then closing user can proceed.
2121 * Starting the timer is contrary to the
2122 * specification, but if we don't get a FIN
2123 * we'll hang forever.
2124 *
2125 * XXXjl:
2126 * we should release the tp also, and use a
2127 * compressed state.
2128 */
2129 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2130 int timeout;
2131
2132 soisdisconnected(so);
2133 timeout = (tcp_fast_finwait2_recycle) ?
2134 tcp_finwait2_timeout : tcp_maxidle;
2135 tcp_timer_activate(tp, TT_2MSL, timeout);
2136 }
2137 tp->t_state = TCPS_FIN_WAIT_2;
2138 }
2139 break;
2140
2141 /*
2142 * In CLOSING STATE in addition to the processing for
2143 * the ESTABLISHED state if the ACK acknowledges our FIN
2144 * then enter the TIME-WAIT state, otherwise ignore
2145 * the segment.
2146 */
2147 case TCPS_CLOSING:
2148 if (ourfinisacked) {
2149 KASSERT(headlocked, ("%s: process_ACK: "
2150 "head not locked", __func__));
2151 tcp_twstart(tp);
2152 INP_INFO_WUNLOCK(&tcbinfo);
2153 headlocked = 0;
2154 m_freem(m);
2155 return;
2156 }
2157 break;
2158
2159 /*
2160 * In LAST_ACK, we may still be waiting for data to drain
2161 * and/or to be acked, as well as for the ack of our FIN.
2162 * If our FIN is now acknowledged, delete the TCB,
2163 * enter the closed state and return.
2164 */
2165 case TCPS_LAST_ACK:
2166 if (ourfinisacked) {
2167 KASSERT(headlocked, ("%s: process_ACK: "
2168 "tcp_close: head not locked", __func__));
2169 tp = tcp_close(tp);
2170 goto drop;
2171 }
2172 break;
2173 }
2174 }
2175
2176 step6:
2177 KASSERT(headlocked, ("%s: step6: head not locked", __func__));
2178 INP_WLOCK_ASSERT(tp->t_inpcb);
2179
2180 /*
2181 * Update window information.
2182 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2183 */
2184 if ((thflags & TH_ACK) &&
2185 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2186 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2187 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2188 /* keep track of pure window updates */
2189 if (tlen == 0 &&
2190 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2191 tcpstat.tcps_rcvwinupd++;
2192 tp->snd_wnd = tiwin;
2193 tp->snd_wl1 = th->th_seq;
2194 tp->snd_wl2 = th->th_ack;
2195 if (tp->snd_wnd > tp->max_sndwnd)
2196 tp->max_sndwnd = tp->snd_wnd;
2197 needoutput = 1;
2198 }
2199
2200 /*
2201 * Process segments with URG.
2202 */
2203 if ((thflags & TH_URG) && th->th_urp &&
2204 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2205 /*
2206 * This is a kludge, but if we receive and accept
2207 * random urgent pointers, we'll crash in
2208 * soreceive. It's hard to imagine someone
2209 * actually wanting to send this much urgent data.
2210 */
2211 SOCKBUF_LOCK(&so->so_rcv);
2212 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2213 th->th_urp = 0; /* XXX */
2214 thflags &= ~TH_URG; /* XXX */
2215 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2216 goto dodata; /* XXX */
2217 }
2218 /*
2219 * If this segment advances the known urgent pointer,
2220 * then mark the data stream. This should not happen
2221 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2222 * a FIN has been received from the remote side.
2223 * In these states we ignore the URG.
2224 *
2225 * According to RFC961 (Assigned Protocols),
2226 * the urgent pointer points to the last octet
2227 * of urgent data. We continue, however,
2228 * to consider it to indicate the first octet
2229 * of data past the urgent section as the original
2230 * spec states (in one of two places).
2231 */
2232 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2233 tp->rcv_up = th->th_seq + th->th_urp;
2234 so->so_oobmark = so->so_rcv.sb_cc +
2235 (tp->rcv_up - tp->rcv_nxt) - 1;
2236 if (so->so_oobmark == 0)
2237 so->so_rcv.sb_state |= SBS_RCVATMARK;
2238 sohasoutofband(so);
2239 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2240 }
2241 SOCKBUF_UNLOCK(&so->so_rcv);
2242 /*
2243 * Remove out of band data so doesn't get presented to user.
2244 * This can happen independent of advancing the URG pointer,
2245 * but if two URG's are pending at once, some out-of-band
2246 * data may creep in... ick.
2247 */
2248 if (th->th_urp <= (u_long)tlen &&
2249 !(so->so_options & SO_OOBINLINE)) {
2250 /* hdr drop is delayed */
2251 tcp_pulloutofband(so, th, m, drop_hdrlen);
2252 }
2253 } else {
2254 /*
2255 * If no out of band data is expected,
2256 * pull receive urgent pointer along
2257 * with the receive window.
2258 */
2259 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2260 tp->rcv_up = tp->rcv_nxt;
2261 }
2262 dodata: /* XXX */
2263 KASSERT(headlocked, ("%s: dodata: head not locked", __func__));
2264 INP_WLOCK_ASSERT(tp->t_inpcb);
2265
2266 /*
2267 * Process the segment text, merging it into the TCP sequencing queue,
2268 * and arranging for acknowledgment of receipt if necessary.
2269 * This process logically involves adjusting tp->rcv_wnd as data
2270 * is presented to the user (this happens in tcp_usrreq.c,
2271 * case PRU_RCVD). If a FIN has already been received on this
2272 * connection then we just ignore the text.
2273 */
2274 if ((tlen || (thflags & TH_FIN)) &&
2275 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2276 tcp_seq save_start = th->th_seq;
2277 m_adj(m, drop_hdrlen); /* delayed header drop */
2278 /*
2279 * Insert segment which includes th into TCP reassembly queue
2280 * with control block tp. Set thflags to whether reassembly now
2281 * includes a segment with FIN. This handles the common case
2282 * inline (segment is the next to be received on an established
2283 * connection, and the queue is empty), avoiding linkage into
2284 * and removal from the queue and repetition of various
2285 * conversions.
2286 * Set DELACK for segments received in order, but ack
2287 * immediately when segments are out of order (so
2288 * fast retransmit can work).
2289 */
2290 if (th->th_seq == tp->rcv_nxt &&
2291 LIST_EMPTY(&tp->t_segq) &&
2292 TCPS_HAVEESTABLISHED(tp->t_state)) {
2293 if (DELAY_ACK(tp))
2294 tp->t_flags |= TF_DELACK;
2295 else
2296 tp->t_flags |= TF_ACKNOW;
2297 tp->rcv_nxt += tlen;
2298 thflags = th->th_flags & TH_FIN;
2299 tcpstat.tcps_rcvpack++;
2300 tcpstat.tcps_rcvbyte += tlen;
2301 ND6_HINT(tp);
2302 SOCKBUF_LOCK(&so->so_rcv);
2303 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2304 m_freem(m);
2305 else
2306 sbappendstream_locked(&so->so_rcv, m);
2307 /* NB: sorwakeup_locked() does an implicit unlock. */
2308 sorwakeup_locked(so);
2309 } else {
2310 /*
2311 * XXX: Due to the header drop above "th" is
2312 * theoretically invalid by now. Fortunately
2313 * m_adj() doesn't actually frees any mbufs
2314 * when trimming from the head.
2315 */
2316 thflags = tcp_reass(tp, th, &tlen, m);
2317 tp->t_flags |= TF_ACKNOW;
2318 }
2319 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2320 tcp_update_sack_list(tp, save_start, save_start + tlen);
2321 #if 0
2322 /*
2323 * Note the amount of data that peer has sent into
2324 * our window, in order to estimate the sender's
2325 * buffer size.
2326 * XXX: Unused.
2327 */
2328 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
2329 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2330 else
2331 len = so->so_rcv.sb_hiwat;
2332 #endif
2333 } else {
2334 m_freem(m);
2335 thflags &= ~TH_FIN;
2336 }
2337
2338 /*
2339 * If FIN is received ACK the FIN and let the user know
2340 * that the connection is closing.
2341 */
2342 if (thflags & TH_FIN) {
2343 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2344 socantrcvmore(so);
2345 /*
2346 * If connection is half-synchronized
2347 * (ie NEEDSYN flag on) then delay ACK,
2348 * so it may be piggybacked when SYN is sent.
2349 * Otherwise, since we received a FIN then no
2350 * more input can be expected, send ACK now.
2351 */
2352 if (tp->t_flags & TF_NEEDSYN)
2353 tp->t_flags |= TF_DELACK;
2354 else
2355 tp->t_flags |= TF_ACKNOW;
2356 tp->rcv_nxt++;
2357 }
2358 switch (tp->t_state) {
2359
2360 /*
2361 * In SYN_RECEIVED and ESTABLISHED STATES
2362 * enter the CLOSE_WAIT state.
2363 */
2364 case TCPS_SYN_RECEIVED:
2365 tp->t_starttime = ticks;
2366 /* FALLTHROUGH */
2367 case TCPS_ESTABLISHED:
2368 tp->t_state = TCPS_CLOSE_WAIT;
2369 break;
2370
2371 /*
2372 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2373 * enter the CLOSING state.
2374 */
2375 case TCPS_FIN_WAIT_1:
2376 tp->t_state = TCPS_CLOSING;
2377 break;
2378
2379 /*
2380 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2381 * starting the time-wait timer, turning off the other
2382 * standard timers.
2383 */
2384 case TCPS_FIN_WAIT_2:
2385 KASSERT(headlocked == 1, ("%s: dodata: "
2386 "TCP_FIN_WAIT_2: head not locked", __func__));
2387 tcp_twstart(tp);
2388 INP_INFO_WUNLOCK(&tcbinfo);
2389 return;
2390 }
2391 }
2392 INP_INFO_WUNLOCK(&tcbinfo);
2393 headlocked = 0;
2394 #ifdef TCPDEBUG
2395 if (so->so_options & SO_DEBUG)
2396 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2397 &tcp_savetcp, 0);
2398 #endif
2399
2400 /*
2401 * Return any desired output.
2402 */
2403 if (needoutput || (tp->t_flags & TF_ACKNOW))
2404 (void) tcp_output(tp);
2405
2406 check_delack:
2407 KASSERT(headlocked == 0, ("%s: check_delack: head locked",
2408 __func__));
2409 INP_INFO_UNLOCK_ASSERT(&tcbinfo);
2410 INP_WLOCK_ASSERT(tp->t_inpcb);
2411 if (tp->t_flags & TF_DELACK) {
2412 tp->t_flags &= ~TF_DELACK;
2413 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
2414 }
2415 INP_WUNLOCK(tp->t_inpcb);
2416 return;
2417
2418 dropafterack:
2419 KASSERT(headlocked, ("%s: dropafterack: head not locked", __func__));
2420 /*
2421 * Generate an ACK dropping incoming segment if it occupies
2422 * sequence space, where the ACK reflects our state.
2423 *
2424 * We can now skip the test for the RST flag since all
2425 * paths to this code happen after packets containing
2426 * RST have been dropped.
2427 *
2428 * In the SYN-RECEIVED state, don't send an ACK unless the
2429 * segment we received passes the SYN-RECEIVED ACK test.
2430 * If it fails send a RST. This breaks the loop in the
2431 * "LAND" DoS attack, and also prevents an ACK storm
2432 * between two listening ports that have been sent forged
2433 * SYN segments, each with the source address of the other.
2434 */
2435 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2436 (SEQ_GT(tp->snd_una, th->th_ack) ||
2437 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2438 rstreason = BANDLIM_RST_OPENPORT;
2439 goto dropwithreset;
2440 }
2441 #ifdef TCPDEBUG
2442 if (so->so_options & SO_DEBUG)
2443 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2444 &tcp_savetcp, 0);
2445 #endif
2446 KASSERT(headlocked, ("%s: headlocked should be 1", __func__));
2447 INP_INFO_WUNLOCK(&tcbinfo);
2448 tp->t_flags |= TF_ACKNOW;
2449 (void) tcp_output(tp);
2450 INP_WUNLOCK(tp->t_inpcb);
2451 m_freem(m);
2452 return;
2453
2454 dropwithreset:
2455 KASSERT(headlocked, ("%s: dropwithreset: head not locked", __func__));
2456 INP_INFO_WUNLOCK(&tcbinfo);
2457
2458 if (tp != NULL) {
2459 tcp_dropwithreset(m, th, tp, tlen, rstreason);
2460 INP_WUNLOCK(tp->t_inpcb);
2461 } else
2462 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
2463 return;
2464
2465 drop:
2466 /*
2467 * Drop space held by incoming segment and return.
2468 */
2469 #ifdef TCPDEBUG
2470 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2471 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2472 &tcp_savetcp, 0);
2473 #endif
2474 if (tp != NULL)
2475 INP_WUNLOCK(tp->t_inpcb);
2476 if (headlocked)
2477 INP_INFO_WUNLOCK(&tcbinfo);
2478 m_freem(m);
2479 }
2480
2481 /*
2482 * Issue RST and make ACK acceptable to originator of segment.
2483 * The mbuf must still include the original packet header.
2484 * tp may be NULL.
2485 */
2486 static void
2487 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
2488 int tlen, int rstreason)
2489 {
2490 struct ip *ip;
2491 #ifdef INET6
2492 struct ip6_hdr *ip6;
2493 #endif
2494
2495 if (tp != NULL) {
2496 INP_WLOCK_ASSERT(tp->t_inpcb);
2497 }
2498
2499 /* Don't bother if destination was broadcast/multicast. */
2500 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2501 goto drop;
2502 #ifdef INET6
2503 if (mtod(m, struct ip *)->ip_v == 6) {
2504 ip6 = mtod(m, struct ip6_hdr *);
2505 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2506 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2507 goto drop;
2508 /* IPv6 anycast check is done at tcp6_input() */
2509 } else
2510 #endif
2511 {
2512 ip = mtod(m, struct ip *);
2513 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2514 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2515 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2516 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2517 goto drop;
2518 }
2519
2520 /* Perform bandwidth limiting. */
2521 if (badport_bandlim(rstreason) < 0)
2522 goto drop;
2523
2524 /* tcp_respond consumes the mbuf chain. */
2525 if (th->th_flags & TH_ACK) {
2526 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
2527 th->th_ack, TH_RST);
2528 } else {
2529 if (th->th_flags & TH_SYN)
2530 tlen++;
2531 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2532 (tcp_seq)0, TH_RST|TH_ACK);
2533 }
2534 return;
2535 drop:
2536 m_freem(m);
2537 }
2538
2539 /*
2540 * Parse TCP options and place in tcpopt.
2541 */
2542 static void
2543 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
2544 {
2545 int opt, optlen;
2546
2547 to->to_flags = 0;
2548 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2549 opt = cp[0];
2550 if (opt == TCPOPT_EOL)
2551 break;
2552 if (opt == TCPOPT_NOP)
2553 optlen = 1;
2554 else {
2555 if (cnt < 2)
2556 break;
2557 optlen = cp[1];
2558 if (optlen < 2 || optlen > cnt)
2559 break;
2560 }
2561 switch (opt) {
2562 case TCPOPT_MAXSEG:
2563 if (optlen != TCPOLEN_MAXSEG)
2564 continue;
2565 if (!(flags & TO_SYN))
2566 continue;
2567 to->to_flags |= TOF_MSS;
2568 bcopy((char *)cp + 2,
2569 (char *)&to->to_mss, sizeof(to->to_mss));
2570 to->to_mss = ntohs(to->to_mss);
2571 break;
2572 case TCPOPT_WINDOW:
2573 if (optlen != TCPOLEN_WINDOW)
2574 continue;
2575 if (!(flags & TO_SYN))
2576 continue;
2577 to->to_flags |= TOF_SCALE;
2578 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
2579 break;
2580 case TCPOPT_TIMESTAMP:
2581 if (optlen != TCPOLEN_TIMESTAMP)
2582 continue;
2583 to->to_flags |= TOF_TS;
2584 bcopy((char *)cp + 2,
2585 (char *)&to->to_tsval, sizeof(to->to_tsval));
2586 to->to_tsval = ntohl(to->to_tsval);
2587 bcopy((char *)cp + 6,
2588 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2589 to->to_tsecr = ntohl(to->to_tsecr);
2590 break;
2591 #ifdef TCP_SIGNATURE
2592 /*
2593 * XXX In order to reply to a host which has set the
2594 * TCP_SIGNATURE option in its initial SYN, we have to
2595 * record the fact that the option was observed here
2596 * for the syncache code to perform the correct response.
2597 */
2598 case TCPOPT_SIGNATURE:
2599 if (optlen != TCPOLEN_SIGNATURE)
2600 continue;
2601 to->to_flags |= TOF_SIGNATURE;
2602 to->to_signature = cp + 2;
2603 break;
2604 #endif
2605 case TCPOPT_SACK_PERMITTED:
2606 if (optlen != TCPOLEN_SACK_PERMITTED)
2607 continue;
2608 if (!(flags & TO_SYN))
2609 continue;
2610 if (!tcp_do_sack)
2611 continue;
2612 to->to_flags |= TOF_SACKPERM;
2613 break;
2614 case TCPOPT_SACK:
2615 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
2616 continue;
2617 if (flags & TO_SYN)
2618 continue;
2619 to->to_flags |= TOF_SACK;
2620 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
2621 to->to_sacks = cp + 2;
2622 tcpstat.tcps_sack_rcv_blocks++;
2623 break;
2624 default:
2625 continue;
2626 }
2627 }
2628 }
2629
2630 /*
2631 * Pull out of band byte out of a segment so
2632 * it doesn't appear in the user's data queue.
2633 * It is still reflected in the segment length for
2634 * sequencing purposes.
2635 */
2636 static void
2637 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
2638 int off)
2639 {
2640 int cnt = off + th->th_urp - 1;
2641
2642 while (cnt >= 0) {
2643 if (m->m_len > cnt) {
2644 char *cp = mtod(m, caddr_t) + cnt;
2645 struct tcpcb *tp = sototcpcb(so);
2646
2647 INP_WLOCK_ASSERT(tp->t_inpcb);
2648
2649 tp->t_iobc = *cp;
2650 tp->t_oobflags |= TCPOOB_HAVEDATA;
2651 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2652 m->m_len--;
2653 if (m->m_flags & M_PKTHDR)
2654 m->m_pkthdr.len--;
2655 return;
2656 }
2657 cnt -= m->m_len;
2658 m = m->m_next;
2659 if (m == NULL)
2660 break;
2661 }
2662 panic("tcp_pulloutofband");
2663 }
2664
2665 /*
2666 * Collect new round-trip time estimate
2667 * and update averages and current timeout.
2668 */
2669 static void
2670 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2671 {
2672 int delta;
2673
2674 INP_WLOCK_ASSERT(tp->t_inpcb);
2675
2676 tcpstat.tcps_rttupdated++;
2677 tp->t_rttupdated++;
2678 if (tp->t_srtt != 0) {
2679 /*
2680 * srtt is stored as fixed point with 5 bits after the
2681 * binary point (i.e., scaled by 8). The following magic
2682 * is equivalent to the smoothing algorithm in rfc793 with
2683 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2684 * point). Adjust rtt to origin 0.
2685 */
2686 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2687 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2688
2689 if ((tp->t_srtt += delta) <= 0)
2690 tp->t_srtt = 1;
2691
2692 /*
2693 * We accumulate a smoothed rtt variance (actually, a
2694 * smoothed mean difference), then set the retransmit
2695 * timer to smoothed rtt + 4 times the smoothed variance.
2696 * rttvar is stored as fixed point with 4 bits after the
2697 * binary point (scaled by 16). The following is
2698 * equivalent to rfc793 smoothing with an alpha of .75
2699 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2700 * rfc793's wired-in beta.
2701 */
2702 if (delta < 0)
2703 delta = -delta;
2704 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2705 if ((tp->t_rttvar += delta) <= 0)
2706 tp->t_rttvar = 1;
2707 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2708 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2709 } else {
2710 /*
2711 * No rtt measurement yet - use the unsmoothed rtt.
2712 * Set the variance to half the rtt (so our first
2713 * retransmit happens at 3*rtt).
2714 */
2715 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2716 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2717 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2718 }
2719 tp->t_rtttime = 0;
2720 tp->t_rxtshift = 0;
2721
2722 /*
2723 * the retransmit should happen at rtt + 4 * rttvar.
2724 * Because of the way we do the smoothing, srtt and rttvar
2725 * will each average +1/2 tick of bias. When we compute
2726 * the retransmit timer, we want 1/2 tick of rounding and
2727 * 1 extra tick because of +-1/2 tick uncertainty in the
2728 * firing of the timer. The bias will give us exactly the
2729 * 1.5 tick we need. But, because the bias is
2730 * statistical, we have to test that we don't drop below
2731 * the minimum feasible timer (which is 2 ticks).
2732 */
2733 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2734 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2735
2736 /*
2737 * We received an ack for a packet that wasn't retransmitted;
2738 * it is probably safe to discard any error indications we've
2739 * received recently. This isn't quite right, but close enough
2740 * for now (a route might have failed after we sent a segment,
2741 * and the return path might not be symmetrical).
2742 */
2743 tp->t_softerror = 0;
2744 }
2745
2746 /*
2747 * Determine a reasonable value for maxseg size.
2748 * If the route is known, check route for mtu.
2749 * If none, use an mss that can be handled on the outgoing
2750 * interface without forcing IP to fragment; if bigger than
2751 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2752 * to utilize large mbufs. If no route is found, route has no mtu,
2753 * or the destination isn't local, use a default, hopefully conservative
2754 * size (usually 512 or the default IP max size, but no more than the mtu
2755 * of the interface), as we can't discover anything about intervening
2756 * gateways or networks. We also initialize the congestion/slow start
2757 * window to be a single segment if the destination isn't local.
2758 * While looking at the routing entry, we also initialize other path-dependent
2759 * parameters from pre-set or cached values in the routing entry.
2760 *
2761 * Also take into account the space needed for options that we
2762 * send regularly. Make maxseg shorter by that amount to assure
2763 * that we can send maxseg amount of data even when the options
2764 * are present. Store the upper limit of the length of options plus
2765 * data in maxopd.
2766 *
2767 * In case of T/TCP, we call this routine during implicit connection
2768 * setup as well (offer = -1), to initialize maxseg from the cached
2769 * MSS of our peer.
2770 *
2771 * NOTE that this routine is only called when we process an incoming
2772 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
2773 */
2774 void
2775 tcp_mss_update(struct tcpcb *tp, int offer,
2776 struct hc_metrics_lite *metricptr, int *mtuflags)
2777 {
2778 int mss;
2779 u_long maxmtu;
2780 struct inpcb *inp = tp->t_inpcb;
2781 struct hc_metrics_lite metrics;
2782 int origoffer = offer;
2783 #ifdef INET6
2784 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2785 size_t min_protoh = isipv6 ?
2786 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
2787 sizeof (struct tcpiphdr);
2788 #else
2789 const size_t min_protoh = sizeof(struct tcpiphdr);
2790 #endif
2791
2792 INP_WLOCK_ASSERT(tp->t_inpcb);
2793
2794 /* Initialize. */
2795 #ifdef INET6
2796 if (isipv6) {
2797 maxmtu = tcp_maxmtu6(&inp->inp_inc, mtuflags);
2798 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt;
2799 } else
2800 #endif
2801 {
2802 maxmtu = tcp_maxmtu(&inp->inp_inc, mtuflags);
2803 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2804 }
2805
2806 /*
2807 * No route to sender, stay with default mss and return.
2808 */
2809 if (maxmtu == 0) {
2810 /*
2811 * In case we return early we need to initialize metrics
2812 * to a defined state as tcp_hc_get() would do for us
2813 * if there was no cache hit.
2814 */
2815 if (metricptr != NULL)
2816 bzero(metricptr, sizeof(struct hc_metrics_lite));
2817 return;
2818 }
2819
2820 /* What have we got? */
2821 switch (offer) {
2822 case 0:
2823 /*
2824 * Offer == 0 means that there was no MSS on the SYN
2825 * segment, in this case we use tcp_mssdflt as
2826 * already assigned to t_maxopd above.
2827 */
2828 offer = tp->t_maxopd;
2829 break;
2830
2831 case -1:
2832 /*
2833 * Offer == -1 means that we didn't receive SYN yet.
2834 */
2835 /* FALLTHROUGH */
2836
2837 default:
2838 /*
2839 * Prevent DoS attack with too small MSS. Round up
2840 * to at least minmss.
2841 */
2842 offer = max(offer, tcp_minmss);
2843 }
2844
2845 /*
2846 * rmx information is now retrieved from tcp_hostcache.
2847 */
2848 tcp_hc_get(&inp->inp_inc, &metrics);
2849 if (metricptr != NULL)
2850 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
2851
2852 /*
2853 * If there's a discovered mtu int tcp hostcache, use it
2854 * else, use the link mtu.
2855 */
2856 if (metrics.rmx_mtu)
2857 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
2858 else {
2859 #ifdef INET6
2860 if (isipv6) {
2861 mss = maxmtu - min_protoh;
2862 if (!path_mtu_discovery &&
2863 !in6_localaddr(&inp->in6p_faddr))
2864 mss = min(mss, tcp_v6mssdflt);
2865 } else
2866 #endif
2867 {
2868 mss = maxmtu - min_protoh;
2869 if (!path_mtu_discovery &&
2870 !in_localaddr(inp->inp_faddr))
2871 mss = min(mss, tcp_mssdflt);
2872 }
2873 /*
2874 * XXX - The above conditional (mss = maxmtu - min_protoh)
2875 * probably violates the TCP spec.
2876 * The problem is that, since we don't know the
2877 * other end's MSS, we are supposed to use a conservative
2878 * default. But, if we do that, then MTU discovery will
2879 * never actually take place, because the conservative
2880 * default is much less than the MTUs typically seen
2881 * on the Internet today. For the moment, we'll sweep
2882 * this under the carpet.
2883 *
2884 * The conservative default might not actually be a problem
2885 * if the only case this occurs is when sending an initial
2886 * SYN with options and data to a host we've never talked
2887 * to before. Then, they will reply with an MSS value which
2888 * will get recorded and the new parameters should get
2889 * recomputed. For Further Study.
2890 */
2891 }
2892 mss = min(mss, offer);
2893
2894 /*
2895 * Sanity check: make sure that maxopd will be large
2896 * enough to allow some data on segments even if the
2897 * all the option space is used (40bytes). Otherwise
2898 * funny things may happen in tcp_output.
2899 */
2900 mss = max(mss, 64);
2901
2902 /*
2903 * maxopd stores the maximum length of data AND options
2904 * in a segment; maxseg is the amount of data in a normal
2905 * segment. We need to store this value (maxopd) apart
2906 * from maxseg, because now every segment carries options
2907 * and thus we normally have somewhat less data in segments.
2908 */
2909 tp->t_maxopd = mss;
2910
2911 /*
2912 * origoffer==-1 indicates that no segments were received yet.
2913 * In this case we just guess.
2914 */
2915 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2916 (origoffer == -1 ||
2917 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2918 mss -= TCPOLEN_TSTAMP_APPA;
2919
2920 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2921 if (mss > MCLBYTES)
2922 mss &= ~(MCLBYTES-1);
2923 #else
2924 if (mss > MCLBYTES)
2925 mss = mss / MCLBYTES * MCLBYTES;
2926 #endif
2927 tp->t_maxseg = mss;
2928 }
2929
2930 void
2931 tcp_mss(struct tcpcb *tp, int offer)
2932 {
2933 int rtt, mss;
2934 u_long bufsize;
2935 struct inpcb *inp;
2936 struct socket *so;
2937 struct hc_metrics_lite metrics;
2938 int mtuflags = 0;
2939 #ifdef INET6
2940 int isipv6;
2941 #endif
2942 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
2943
2944 tcp_mss_update(tp, offer, &metrics, &mtuflags);
2945
2946 mss = tp->t_maxseg;
2947 inp = tp->t_inpcb;
2948 #ifdef INET6
2949 isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2950 #endif
2951
2952 /*
2953 * If there's a pipesize, change the socket buffer to that size,
2954 * don't change if sb_hiwat is different than default (then it
2955 * has been changed on purpose with setsockopt).
2956 * Make the socket buffers an integral number of mss units;
2957 * if the mss is larger than the socket buffer, decrease the mss.
2958 */
2959 so = inp->inp_socket;
2960 SOCKBUF_LOCK(&so->so_snd);
2961 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
2962 bufsize = metrics.rmx_sendpipe;
2963 else
2964 bufsize = so->so_snd.sb_hiwat;
2965 if (bufsize < mss)
2966 mss = bufsize;
2967 else {
2968 bufsize = roundup(bufsize, mss);
2969 if (bufsize > sb_max)
2970 bufsize = sb_max;
2971 if (bufsize > so->so_snd.sb_hiwat)
2972 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
2973 }
2974 SOCKBUF_UNLOCK(&so->so_snd);
2975 tp->t_maxseg = mss;
2976
2977 SOCKBUF_LOCK(&so->so_rcv);
2978 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
2979 bufsize = metrics.rmx_recvpipe;
2980 else
2981 bufsize = so->so_rcv.sb_hiwat;
2982 if (bufsize > mss) {
2983 bufsize = roundup(bufsize, mss);
2984 if (bufsize > sb_max)
2985 bufsize = sb_max;
2986 if (bufsize > so->so_rcv.sb_hiwat)
2987 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
2988 }
2989 SOCKBUF_UNLOCK(&so->so_rcv);
2990 /*
2991 * While we're here, check the others too.
2992 */
2993 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
2994 tp->t_srtt = rtt;
2995 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2996 tcpstat.tcps_usedrtt++;
2997 if (metrics.rmx_rttvar) {
2998 tp->t_rttvar = metrics.rmx_rttvar;
2999 tcpstat.tcps_usedrttvar++;
3000 } else {
3001 /* default variation is +- 1 rtt */
3002 tp->t_rttvar =
3003 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3004 }
3005 TCPT_RANGESET(tp->t_rxtcur,
3006 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3007 tp->t_rttmin, TCPTV_REXMTMAX);
3008 }
3009 if (metrics.rmx_ssthresh) {
3010 /*
3011 * There's some sort of gateway or interface
3012 * buffer limit on the path. Use this to set
3013 * the slow start threshhold, but set the
3014 * threshold to no less than 2*mss.
3015 */
3016 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh);
3017 tcpstat.tcps_usedssthresh++;
3018 }
3019 if (metrics.rmx_bandwidth)
3020 tp->snd_bandwidth = metrics.rmx_bandwidth;
3021
3022 /*
3023 * Set the slow-start flight size depending on whether this
3024 * is a local network or not.
3025 *
3026 * Extend this so we cache the cwnd too and retrieve it here.
3027 * Make cwnd even bigger than RFC3390 suggests but only if we
3028 * have previous experience with the remote host. Be careful
3029 * not make cwnd bigger than remote receive window or our own
3030 * send socket buffer. Maybe put some additional upper bound
3031 * on the retrieved cwnd. Should do incremental updates to
3032 * hostcache when cwnd collapses so next connection doesn't
3033 * overloads the path again.
3034 *
3035 * XXXAO: Initializing the CWND from the hostcache is broken
3036 * and in its current form not RFC conformant. It is disabled
3037 * until fixed or removed entirely.
3038 *
3039 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
3040 * We currently check only in syncache_socket for that.
3041 */
3042 /* #define TCP_METRICS_CWND */
3043 #ifdef TCP_METRICS_CWND
3044 if (metrics.rmx_cwnd)
3045 tp->snd_cwnd = max(mss,
3046 min(metrics.rmx_cwnd / 2,
3047 min(tp->snd_wnd, so->so_snd.sb_hiwat)));
3048 else
3049 #endif
3050 if (tcp_do_rfc3390)
3051 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3052 #ifdef INET6
3053 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3054 (!isipv6 && in_localaddr(inp->inp_faddr)))
3055 #else
3056 else if (in_localaddr(inp->inp_faddr))
3057 #endif
3058 tp->snd_cwnd = mss * ss_fltsz_local;
3059 else
3060 tp->snd_cwnd = mss * ss_fltsz;
3061
3062 /* Check the interface for TSO capabilities. */
3063 if (mtuflags & CSUM_TSO)
3064 tp->t_flags |= TF_TSO;
3065 }
3066
3067 /*
3068 * Determine the MSS option to send on an outgoing SYN.
3069 */
3070 int
3071 tcp_mssopt(struct in_conninfo *inc)
3072 {
3073 int mss = 0;
3074 u_long maxmtu = 0;
3075 u_long thcmtu = 0;
3076 size_t min_protoh;
3077
3078 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3079
3080 #ifdef INET6
3081 if (inc->inc_flags & INC_ISIPV6) {
3082 mss = tcp_v6mssdflt;
3083 maxmtu = tcp_maxmtu6(inc, NULL);
3084 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3085 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3086 } else
3087 #endif
3088 {
3089 mss = tcp_mssdflt;
3090 maxmtu = tcp_maxmtu(inc, NULL);
3091 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3092 min_protoh = sizeof(struct tcpiphdr);
3093 }
3094 if (maxmtu && thcmtu)
3095 mss = min(maxmtu, thcmtu) - min_protoh;
3096 else if (maxmtu || thcmtu)
3097 mss = max(maxmtu, thcmtu) - min_protoh;
3098
3099 return (mss);
3100 }
3101
3102
3103 /*
3104 * On a partial ack arrives, force the retransmission of the
3105 * next unacknowledged segment. Do not clear tp->t_dupacks.
3106 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3107 * be started again.
3108 */
3109 static void
3110 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3111 {
3112 tcp_seq onxt = tp->snd_nxt;
3113 u_long ocwnd = tp->snd_cwnd;
3114
3115 INP_WLOCK_ASSERT(tp->t_inpcb);
3116
3117 tcp_timer_activate(tp, TT_REXMT, 0);
3118 tp->t_rtttime = 0;
3119 tp->snd_nxt = th->th_ack;
3120 /*
3121 * Set snd_cwnd to one segment beyond acknowledged offset.
3122 * (tp->snd_una has not yet been updated when this function is called.)
3123 */
3124 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3125 tp->t_flags |= TF_ACKNOW;
3126 (void) tcp_output(tp);
3127 tp->snd_cwnd = ocwnd;
3128 if (SEQ_GT(onxt, tp->snd_nxt))
3129 tp->snd_nxt = onxt;
3130 /*
3131 * Partial window deflation. Relies on fact that tp->snd_una
3132 * not updated yet.
3133 */
3134 if (tp->snd_cwnd > th->th_ack - tp->snd_una)
3135 tp->snd_cwnd -= th->th_ack - tp->snd_una;
3136 else
3137 tp->snd_cwnd = 0;
3138 tp->snd_cwnd += tp->t_maxseg;
3139 }
Cache object: a7e78000489779d6f25ce5ae65946266
|